import gradio as gr
import os
import time
import json
import re
from uuid import uuid4
from datetime import datetime
from duckduckgo_search import DDGS
from sentence_transformers import SentenceTransformer, util
from typing import List, Dict, Any, Optional, Union, Tuple
import logging
import numpy as np
from collections import deque
from huggingface_hub import InferenceClient
import requests
import arxiv
import scholarly
import pymed
import wikipedia
import trafilatura
from trafilatura import extract, fetch_url
import pickle
import faiss
import threading
from concurrent.futures import ThreadPoolExecutor, as_completed
import tiktoken

logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger(name)

HF_API_KEY = os.environ.get("HF_API_KEY")
if not HF_API_KEY:
    raise ValueError("Please set the HF_API_KEY environment variable.")

client = InferenceClient(provider="hf-inference", api_key=HF_API_KEY)

# Enhanced Model Configuration
MAIN_LLM_MODEL = "mistralai/Mistral-Nemo-Instruct-2407"  # Most powerful for main reasoning
REASONING_LLM_MODEL = "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B"  # Specialized for analytical tasks
CRITIC_LLM_MODEL = "Qwen/QwQ-32B-Preview"  # Diverse perspective for critiques
SPECIALIST_MODELS = {
    "medical": "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B",
    "scientific": "mistralai/Mistral-7B-Instruct-v0.2",
    "financial": "mistralai/Mistral-7B-Instruct-v0.3",
    "legal": "Qwen/Qwen2.5-Coder-32B-Instruct"
}
ENSEMBLE_MODELS = [MAIN_LLM_MODEL, REASONING_LLM_MODEL, CRITIC_LLM_MODEL] + list(SPECIALIST_MODELS.values())

# Enhanced Parameters
MAX_ITERATIONS = 100  # Increased for deeper research
TIMEOUT = 300  # Longer timeout
RETRY_DELAY = 15
NUM_RESULTS = 50  # More comprehensive search
SIMILARITY_THRESHOLD = 0.12  # More lenient for broader coverage
MAX_CONTEXT_ITEMS = 100
MAX_HISTORY_ITEMS = 25
MAX_FULL_TEXT_LENGTH = 50000
FAISS_INDEX_PATH = "research_index.faiss"
RESEARCH_DATA_PATH = "research_data.pkl"
PAPER_SUMMARIES_PATH = "paper_summaries.pkl" #New path for storing paper summary

try:
main_similarity_model = SentenceTransformer('sentence-transformers/all-mpnet-base-v2')
concept_similarity_model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
document_similarity_model = SentenceTransformer('sentence-transformers/multi-qa-mpnet-base-dot-v1')

embedding_dim = document_similarity_model.get_sentence_embedding_dimension()
if os.path.exists(FAISS_INDEX_PATH):
    index = faiss.read_index(FAISS_INDEX_PATH)
    logger.info(f"Loaded FAISS index from {FAISS_INDEX_PATH}")
else:
    index = faiss.IndexFlatIP(embedding_dim)  # Use IndexFlatIP for inner product (cosine similarity).
    logger.info("Created a new FAISS index.")
content_copy
download
 Use code with caution.

except Exception as e:
logger.error(f"Failed to load models or initialize FAISS: {e}")
raise

def get_token_count(text):
try:
encoding = tiktoken.get_encoding("cl100k_base")
return len(encoding.encode(text))
except:
return len(text.split()) * 1.3

def save_research_data(data, index):
try:
with open(RESEARCH_DATA_PATH, "wb") as f:
pickle.dump(data, f)
faiss.write_index(index, FAISS_INDEX_PATH)
logger.info(f"Research data and index saved to {RESEARCH_DATA_PATH} and {FAISS_INDEX_PATH}")
except Exception as e:
logger.error(f"Error saving research data: {e}")

def load_research_data():
if os.path.exists(RESEARCH_DATA_PATH):
try:
with open(RESEARCH_DATA_PATH, "rb") as f:
data = pickle.load(f)
logger.info(f"Loaded research data from {RESEARCH_DATA_PATH}")
return data
except Exception as e:
logger.error(f"Error loading research data: {e}")
return {}
else:
logger.info("No existing research data found.")
return {}

def save_paper_summaries(summaries: Dict[str, str]):
try:
with open(PAPER_SUMMARIES_PATH, "wb") as f:
pickle.dump(summaries, f)
logger.info(f"Paper summaries saved to {PAPER_SUMMARIES_PATH}")
except Exception as e:
logger.error(f"Error saving paper summaries: {e}")

def load_paper_summaries() -> Dict[str, str]:
if os.path.exists(PAPER_SUMMARIES_PATH):
try:
with open(PAPER_SUMMARIES_PATH, "rb") as f:
data = pickle.load(f)
logger.info(f"Loaded paper summaries from {PAPER_SUMMARIES_PATH}")
return data
except Exception as e:
logger.error(f"Error loading paper summaries: {e}")
return {}
else:
logger.info("No existing paper summaries found.")
return {}

def hf_inference(model_name, prompt, max_tokens=2000, retries=5, stream=False):  # Added stream parameter
for attempt in range(retries):
try:
messages = [{"role": "user", "content": prompt}]
response_generator = client.chat.completions.create(
model=model_name,
messages=messages,
max_tokens=max_tokens,
stream=stream  # Pass the stream parameter
)
if stream:
return response_generator # Return the generator directly
else:
# If not streaming, get the full response
response = next(response_generator) # Consume the first chunk to get complete object
return {"generated_text": response.choices[0].message.content}
except Exception as e:
if attempt == retries - 1:
logger.error(f"Request failed after {retries} retries: {e}")
return {"error": f"Request failed after {retries} retries: {e}"}
time.sleep(RETRY_DELAY * (1 + attempt))
return {"error": "Request failed after multiple retries."}

def ensemble_inference(prompt, models=ENSEMBLE_MODELS, max_tokens=1500, stream=False): #Added stream
results = []

if stream: # If streaming, return a generator that yields from each model
    def generate_responses():
        with ThreadPoolExecutor(max_workers=len(models)) as executor:
            futures = {executor.submit(hf_inference, model, prompt, max_tokens, stream=True): model for model in models}

            for future in as_completed(futures):
                model = future_to_model[future]
                try:
                    for chunk in future.result():  # Iterate through chunks
                        yield {"model": model, "text": chunk.choices[0].delta.content} #yield the content of the chunk
                except Exception as e:
                    logger.error(f"Error with model {model}: {e}")
                    yield {"model": model, "text": f"Error: {e}"}
    return generate_responses() # return the generator

else: #Non-streaming behavior
    with ThreadPoolExecutor(max_workers=len(models)) as executor:
        future_to_model = {executor.submit(hf_inference, model, prompt, max_tokens, stream=False): model for model in models}
        for future in as_completed(future_to_model):
            model = future_to_model[future]
            try:
                result = future.result()
                if "generated_text" in result:
                    results.append({"model": model, "text": result["generated_text"]})
            except Exception as e:
                logger.error(f"Error with model {model}: {e}")

    if not results:
        return {"error": "All models failed to generate responses"}

    if len(results) == 1:
        return {"generated_text": results[0]["text"]}

    synthesis_prompt = "Synthesize these expert responses into a single coherent answer:\n\n"
    for result in results:
        synthesis_prompt += f"Expert {results.index(result) + 1} ({result['model'].split('/')[-1]}):\n{result['text']}\n\n"

    synthesis = hf_inference(MAIN_LLM_MODEL, synthesis_prompt) # Use a consistent model for final synthesis
    if "generated_text" in synthesis:
        return synthesis
    else:
        return {"generated_text": max(results, key=lambda x: len(x["text"]))["text"]} # Fallback
content_copy
download
 Use code with caution.

def tool_search_web(query: str, num_results: int = NUM_RESULTS, safesearch: str = "moderate",
time_filter: Optional[str] = None, region: str = "wt-wt", language: str = "en-us") -> list:
try:
with DDGS() as ddgs:
kwargs = {
"keywords": query,
"max_results": num_results,
"safesearch": safesearch,
"region": region,
"hreflang": language,
}
if time_filter:
if time_filter in ['d', 'w', 'm', 'y']:
kwargs["time"] = time_filter

results = [r for r in ddgs.text(**kwargs)]
        if results:
            return [{"title": r["title"], "snippet": r["body"], "url": r["href"]} for r in results]
        else:
            if time_filter and "time" in kwargs:
                del kwargs["time"]
                results = [r for r in ddgs.text(**kwargs)]
                if results:
                    return [{"title": r["title"], "snippet": r["body"], "url": r["href"]} for r in results]
            return []
except Exception as e:
    logger.error(f"DuckDuckGo search error: {e}")
    return []
content_copy
download
 Use code with caution.

def tool_search_arxiv(query: str, max_results: int = 5) -> list:
try:
client = arxiv.Client()
search = arxiv.Search(
query=query,
max_results=max_results,
sort_by=arxiv.SortCriterion.Relevance
)
results = []
for paper in client.results(search):
results.append({
"title": paper.title,
"snippet": paper.summary[:500] + "..." if len(paper.summary) > 500 else paper.summary,
"url": paper.pdf_url,
"authors": ", ".join(author.name for author in paper.authors),
"published": paper.published.strftime("%Y-%m-%d") if paper.published else "Unknown",
"source": "arXiv"
})
return results
except Exception as e:
logger.error(f"arXiv search error: {e}")
return []

def tool_search_pubmed(query: str, max_results: int = 5) -> list:
try:
pubmed = pymed.PubMed(tool="ResearchAssistant", email="researcher@example.com")
results = list(pubmed.query(query, max_results=max_results))

output = []
    for article in results:
        try:
            data = article.toDict()
            output.append({
                "title": data.get("title", "No title"),
                "snippet": data.get("abstract", "No abstract")[:500] + "..." if data.get("abstract", "") and len(data.get("abstract", "")) > 500 else data.get("abstract", "No abstract"),
                "url": f"https://pubmed.ncbi.nlm.nih.gov/{data.get('pubmed_id')}/",
                "authors": ", ".join(author.get("name", "") for author in data.get("authors", [])),
                "published": data.get("publication_date", "Unknown"),
                "source": "PubMed"
            })
        except:
            continue
    return output
except Exception as e:
    logger.error(f"PubMed search error: {e}")
    return []
content_copy
download
 Use code with caution.

def tool_search_wikipedia(query: str, max_results: int = 3) -> list:
try:
search_results = wikipedia.search(query, results=max_results)
results = []

for title in search_results:
        try:
            page = wikipedia.page(title)
            summary = page.summary
            snippet = summary[:500] + "..." if len(summary) > 500 else summary
            results.append({
                "title": page.title,
                "snippet": snippet,
                "url": page.url,
                "source": "Wikipedia"
            })
        except (wikipedia.exceptions.DisambiguationError, wikipedia.exceptions.PageError):
            continue

    return results
except Exception as e:
    logger.error(f"Wikipedia search error: {e}")
    return []
content_copy
download
 Use code with caution.

def tool_search_scholar(query: str, max_results: int = 5) -> list:
try:
search_query = scholarly.search_pubs(query)
results = []
for _ in range(max_results):
try:
result = next(search_query)
results.append({
"title": result.get("bib", {}).get("title", "No title"),
"snippet": result.get("bib", {}).get("abstract", "No abstract")[:500] + "..." if result.get("bib", {}).get("abstract") else result.get("bib", {}).get("abstract", "No abstract"),
"url": result.get("pub_url", "#"),
"authors": ", ".join(result.get("bib", {}).get("author", [])),
"published": result.get("bib", {}).get("pub_year", "Unknown"),
"source": "Google Scholar"
})
except StopIteration:
break
except Exception as e:
logger.warning(f"Error processing Scholar result: {e}")
continue
return results
except Exception as e:
logger.error(f"Google Scholar search error: {e}")
return []

def extract_article_content(url: str) -> str:
try:
downloaded = fetch_url(url)
if downloaded is None:
return ""
return extract(downloaded, favor_precision=True)
except Exception as e:
logger.error(f"Failed to extract article content from {url}: {e}")
return ""

def tool_reason(prompt: str, search_results: list, reasoning_context: list = [],
critique: str = "", focus_areas: list = []) -> str:
if not search_results:
return "No search results to reason about."

reasoning_input = "Reason about the following search results in relation to the prompt:\n\n"
reasoning_input += f"Prompt: {prompt}\n\n"

if focus_areas:
    reasoning_input += f"Focus particularly on these aspects: {', '.join(focus_areas)}\n\n"

results_by_source = {}
for i, result in enumerate(search_results):
    source = result.get('source', 'Web Search') # Default to 'Web Search'
    if source not in results_by_source:
        results_by_source[source] = []
    results_by_source[source].append((i, result))

for source, results in results_by_source.items():
    reasoning_input += f"\n--- {source} Results ---\n"
    for i, result in results:
        reasoning_input += f"- Result {i + 1}: Title: {result['title']}\n  Snippet: {result['snippet']}\n"
        if 'authors' in result:
            reasoning_input += f"  Authors: {result['authors']}\n"
        if 'published' in result:
            reasoning_input += f"  Published: {result['published']}\n"
        reasoning_input += "\n"

if reasoning_context:
    recent_context = reasoning_context[-MAX_HISTORY_ITEMS:]  # Limit history
    reasoning_input += "\nPrevious Reasoning Context:\n" + "\n".join(recent_context)

if critique:
    reasoning_input += f"\n\nRecent critique to address: {critique}\n"

reasoning_input += "\nProvide a thorough, nuanced analysis that builds upon previous reasoning if applicable. Consider multiple perspectives, potential contradictions in the search results, and the reliability of different sources.  Address any specific critiques."

reasoning_output = ensemble_inference(reasoning_input) # Use ensemble for high-quality reasoning.

if isinstance(reasoning_output, dict) and "generated_text" in reasoning_output:
    return reasoning_output["generated_text"].strip()
else:
    logger.error(f"Failed to generate reasoning: {reasoning_output}")
    return "Could not generate reasoning due to an error."
content_copy
download
 Use code with caution.

def tool_summarize(insights: list, prompt: str, contradictions: list = []) -> str:
if not insights:
return "No insights to summarize."

summarization_input = f"Synthesize the following insights into a cohesive and comprehensive summary regarding: '{prompt}'\n\n"

max_tokens = 12000  # Increased token limit
selected_insights = []
token_count = get_token_count(summarization_input) + get_token_count("\n\n".join(contradictions))

for insight in reversed(insights):
    insight_tokens = get_token_count(insight)
    if token_count + insight_tokens < max_tokens:
        selected_insights.insert(0, insight)
        token_count += insight_tokens
    else:
        break

summarization_input += "\n\n".join(selected_insights)

if contradictions:
    summarization_input += "\n\nAddress these specific contradictions:\n" + "\n".join(contradictions)

summarization_input += "\n\nProvide a well-structured summary that:\n1. Presents the main findings\n2. Acknowledges limitations and uncertainties\n3. Highlights areas of consensus and disagreement\n4. Suggests potential directions for further inquiry\n5. Evaluates the strength of evidence for key claims"

summarization_output = ensemble_inference(summarization_input)

if isinstance(summarization_output, dict) and "generated_text" in summarization_output:
    return summarization_output["generated_text"].strip()
else:
    logger.error(f"Failed to generate summary: {summarization_output}")
    return "Could not generate a summary due to an error."
content_copy
download
 Use code with caution.

def tool_generate_search_query(prompt: str, previous_queries: list = [],
failed_queries: list = [], focus_areas: list = []) -> str:
query_gen_input = f"Generate an effective search query for the following prompt: {prompt}\n"

if previous_queries:
    recent_queries = previous_queries[-MAX_HISTORY_ITEMS:]
    query_gen_input += "Previous search queries:\n" + "\n".join(recent_queries) + "\n"

if failed_queries:
    query_gen_input += "These queries didn't yield useful results:\n" + "\n".join(failed_queries) + "\n"

if focus_areas:
    query_gen_input += f"Focus particularly on these aspects: {', '.join(focus_areas)}\n"

query_gen_input += "Refine the search query based on previous queries, aiming for more precise results. Consider using advanced search operators like site:, filetype:, intitle:, etc. when appropriate. Make sure the query is well-formed for academic and scientific search engines.\n"
query_gen_input += "Search Query:"

query_gen_output = hf_inference(MAIN_LLM_MODEL, query_gen_input)

if isinstance(query_gen_output, dict) and 'generated_text' in query_gen_output:
    return query_gen_output['generated_text'].strip()

logger.error(f"Failed to generate search query: {query_gen_output}")
return ""
content_copy
download
 Use code with caution.

def tool_critique_reasoning(reasoning_output: str, prompt: str,
previous_critiques: list = []) -> str:
critique_input = f"Critically evaluate the following reasoning output in relation to the prompt:\n\nPrompt: {prompt}\n\nReasoning: {reasoning_output}\n\n"

if previous_critiques:
    critique_input += "Previous critiques that should be addressed:\n" + "\n".join(previous_critiques[-MAX_HISTORY_ITEMS:]) + "\n\n"

critique_input += "Identify any flaws, biases, logical fallacies, unsupported claims, or areas for improvement. Be specific and constructive. Suggest concrete ways to enhance the reasoning. Also evaluate the strength of evidence and whether conclusions are proportionate to the available information."

critique_output = hf_inference(CRITIC_LLM_MODEL, critique_input) # Use specialized critique model.

if isinstance(critique_output, dict) and "generated_text" in critique_output:
    return critique_output["generated_text"].strip()

logger.error(f"Failed to generate critique: {critique_output}")
return "Could not generate a critique due to an error."
content_copy
download
 Use code with caution.

def tool_identify_contradictions(insights: list) -> list:
if len(insights) < 2:
return []

max_tokens = 12000  # Increased token limit for potentially more contradictions
selected_insights = []
token_count = 0

for insight in reversed(insights):
    insight_tokens = get_token_count(insight)
    if token_count + insight_tokens < max_tokens:
        selected_insights.insert(0, insight)
        token_count += insight_tokens
    else:
        break

contradiction_input = "Identify specific contradictions in these insights:\n\n" + "\n\n".join(selected_insights)
contradiction_input += "\n\nList each contradiction as a separate numbered point. For each contradiction, cite the specific claims that are in tension and evaluate which claim is better supported. If no contradictions exist, respond with 'No contradictions found.'"

contradiction_output = hf_inference(CRITIC_LLM_MODEL, contradiction_input)  # Use critique model

if isinstance(contradiction_output, dict) and "generated_text" in contradiction_output:
    result = contradiction_output["generated_text"].strip()
    if result == "No contradictions found.":
        return []
    # More robust contradiction extraction, handles multi-sentence contradictions
    contradictions = re.findall(r'\d+\.\s+(.*?)(?=\d+\.|$)', result, re.DOTALL)
    return [c.strip() for c in contradictions if c.strip()]

logger.error(f"Failed to identify contradictions: {contradiction_output}")
return []
content_copy
download
 Use code with caution.

def tool_identify_focus_areas(prompt: str, insights: list = [],
failed_areas: list = []) -> list:
focus_input = f"Based on this research prompt: '{prompt}'\n\n"

if insights:
    recent_insights = insights[-5:] if len(insights) > 5 else insights
    focus_input += "And these existing insights:\n" + "\n".join(recent_insights) + "\n\n"

if failed_areas:
    focus_input += f"These focus areas didn't yield useful results: {', '.join(failed_areas)}\n\n"

focus_input += "Identify 3-5 specific aspects that should be investigated further to get a complete understanding. Be precise and prioritize underexplored areas. For each suggested area, briefly explain why it's important to investigate."

focus_output = hf_inference(MAIN_LLM_MODEL, focus_input)  # Consistent model

if isinstance(focus_output, dict) and "generated_text" in focus_output:
    result = focus_output["generated_text"].strip()
    # More robust extraction, handles different list formats
    areas = re.findall(r'(?:^|\n)(?:\d+\.|\*|\-)\s*(.*?)(?=(?:\n(?:\d+\.|\*|\-|$))|$)', result)
    return [area.strip() for area in areas if area.strip()][:5]

logger.error(f"Failed to identify focus areas: {focus_output}")
return []
content_copy
download
 Use code with caution.

def add_to_faiss_index(text: str):
embedding = document_similarity_model.encode(text, convert_to_tensor=True)
embedding_np = embedding.cpu().numpy().reshape(1, -1)
if embedding_np.shape[1] != embedding_dim:
logger.error(f"Embedding dimension mismatch: expected {embedding_dim}, got {embedding_np.shape[1]}")
return
faiss.normalize_L2(embedding_np)  # Normalize for cosine similarity.
index.add(embedding_np)

def search_faiss_index(query: str, top_k: int = 5) -> List[str]:
query_embedding = document_similarity_model.encode(query, convert_to_tensor=True)
query_embedding_np = query_embedding.cpu().numpy().reshape(1, -1)
faiss.normalize_L2(query_embedding_np)
distances, indices = index.search(query_embedding_np, top_k)
return indices[0].tolist()

def filter_results(search_results, prompt, previous_snippets=None):
if not main_similarity_model or not search_results:
return search_results

try:
    prompt_embedding = main_similarity_model.encode(prompt, convert_to_tensor=True)
    filtered_results = []

    seen_snippets = set()
    if previous_snippets:
        seen_snippets.update(previous_snippets)

    for result in search_results:
        combined_text = result['title'] + " " + result['snippet']

        if result['snippet'] in seen_snippets:  # Prevent exact duplicates
            continue

        result_embedding = main_similarity_model.encode(combined_text, convert_to_tensor=True)
        cosine_score = util.pytorch_cos_sim(prompt_embedding, result_embedding)[0][0].item()
        
        if cosine_score >= SIMILARITY_THRESHOLD:
            result['relevance_score'] = cosine_score
            filtered_results.append(result)
            seen_snippets.add(result['snippet']) # Add snippets after filtering
            add_to_faiss_index(result['snippet'])


    filtered_results.sort(key=lambda x: x.get('relevance_score', 0), reverse=True)  # Sort by relevance.
    return filtered_results

except Exception as e:
    logger.error(f"Error during filtering: {e}")
    return search_results  # Return original results on error.
content_copy
download
 Use code with caution.

def tool_extract_key_entities(prompt: str) -> list:
entity_input = f"Extract the key entities (people, organizations, concepts, technologies, events, time periods, locations, etc.) from this research prompt that should be investigated individually:\n\n{prompt}\n\nList the 5-7 most important entities, one per line, with a brief explanation (2-3 sentences) of why each is central to the research question."

entity_output = hf_inference(MAIN_LLM_MODEL, entity_input)

if isinstance(entity_output, dict) and "generated_text" in entity_output:
    result = entity_output["generated_text"].strip()
    entities = [e.strip() for e in result.split('\n') if e.strip()]
    return entities[:7] # Limit to top 7 entities

logger.error(f"Failed to extract key entities: {entity_output}")
return []
content_copy
download
 Use code with caution.

def tool_meta_analyze(entity_insights: Dict[str, list], prompt: str) -> str:
if not entity_insights:
return "No entity insights to analyze."

meta_input = f"Perform a meta-analysis across these different entities related to the prompt: '{prompt}'\n\n"

for entity, insights in entity_insights.items():
    if insights:
        meta_input += f"\n--- {entity} ---\n" + insights[-1] + "\n"  # Most recent insight for each entity

meta_input += "\nProvide a high-level synthesis that identifies:\n1. Common themes across entities\n2. Important differences and contradictions\n3. How these entities interact or influence each other\n4. The broader implications for the original research question\n5. A systems-level understanding of how these elements fit together"

meta_output = ensemble_inference(meta_input)  # Ensemble for meta-analysis

if isinstance(meta_output, dict) and "generated_text" in meta_output:
    return meta_output["generated_text"].strip()

logger.error(f"Failed to perform meta-analysis: {meta_output}")
return "Could not generate a meta-analysis due to an error."
content_copy
download
 Use code with caution.

def tool_draft_research_plan(prompt: str, entities: list, focus_areas: list = []) -> str:
plan_input = f"Create a detailed research plan for investigating this question: '{prompt}'\n\n"

if entities:
    plan_input += "Key entities to investigate:\n" + "\n".join(entities) + "\n\n"

if focus_areas:
    plan_input += "Additional focus areas:\n" + "\n".join(focus_areas) + "\n\n"

plan_input += "The research plan should include:\n"
plan_input += "1. Main research questions and sub-questions\n"
plan_input += "2. Methodology for investigating each aspect\n"
plan_input += "3. Potential sources and databases to consult\n"
plan_input += "4. Suggested sequence of investigation\n"
plan_input += "5. Potential challenges and how to address them\n"
plan_input += "6. Criteria for evaluating the quality of findings"

plan_output = hf_inference(REASONING_LLM_MODEL, plan_input) # Use reasoning model

if isinstance(plan_output, dict) and "generated_text" in plan_output:
    return plan_output["generated_text"].strip()

logger.error(f"Failed to generate research plan: {plan_output}")
return "Could not generate a research plan due to an error."
content_copy
download
 Use code with caution.

def tool_extract_article(url: str) -> str:
extracted_text = extract_article_content(url)
return extracted_text if extracted_text else f"Could not extract content from {url}"

New tool for summarizing a single paper

def tool_summarize_paper(paper_text: str) -> str:
summarization_prompt = f"""Summarize this academic paper, focusing on the following:

Main Research Question(s): What questions does the paper address?

Methodology: Briefly describe the methods used (e.g., experiments, surveys, simulations, theoretical analysis).

Key Findings: What are the most important results or conclusions?

Limitations:  What are the acknowledged limitations of the study?

Implications: What are the broader implications of the findings, according to the authors?
Paper Text:
{paper_text[:MAX_FULL_TEXT_LENGTH]}
"""  # Truncate if necessary
summary = hf_inference(REASONING_LLM_MODEL, summarization_prompt, max_tokens=500)

if isinstance(summary, dict) and "generated_text" in summary:
return summary["generated_text"].strip()
else:
logger.error(f"Failed to generate summary: {summary}")
return "Could not generate a summary due to an error."

def tool_search_patents(query: str, max_results: int = 10) -> list:
    """Search patent databases including USPTO and EPO"""
    # Implementation details...

def tool_search_clinical_trials(query: str, max_results: int = 10) -> list:
    """Search ClinicalTrials.gov and WHO ICTRP"""
    # Implementation details...

def tool_search_datasets(query: str, max_results: int = 10) -> list:
    """Search academic datasets from repositories like Kaggle, UCI, etc."""
    # Implementation details...

def tool_search_conferences(query: str, max_results: int = 10) -> list:
    """Search major conference proceedings"""
    # Implementation details...

tools = {
"search_web": {
"function": tool_search_web,
"description": "Searches the web for information.",
"parameters": {
"query": {"type": "string", "description": "The search query."},
"num_results": {"type": "integer", "description": "Number of results to return."},
"time_filter": {"type": "string", "description": "Optional time filter (d, w, m, y)."},
"region": {"type": "string", "description": "Optional region code."},
"language": {"type": "string", "description": "Optional language code."}
},
},
"search_arxiv": {
"function": tool_search_arxiv,
"description": "Searches arXiv for scientific papers.",
"parameters": {
"query": {"type": "string", "description": "The search query for scientific papers."},
"max_results": {"type": "integer", "description": "Maximum number of papers to return."}
},
},
"search_pubmed": {
"function": tool_search_pubmed,
"description": "Searches PubMed for medical and scientific literature.",
"parameters": {
"query": {"type": "string", "description": "The search query for medical literature."},
"max_results": {"type": "integer", "description": "Maximum number of articles to return."}
},
},
"search_wikipedia": {
"function": tool_search_wikipedia,
"description": "Searches Wikipedia for information.",
"parameters": {
"query": {"type": "string", "description": "The search query for Wikipedia."},
"max_results": {"type": "integer", "description": "Maximum number of articles to return."}
},
},
"search_scholar": {
"function": tool_search_scholar,
"description": "Searches Google Scholar for academic publications.",
"parameters": {
"query": {"type": "string", "description": "The search query for Google Scholar."},
"max_results": {"type": "integer", "description": "Maximum number of articles to return."}
}
},
"extract_article": {
"function": tool_extract_article,
"description": "Extracts the main content from a web article URL",
"parameters": {
"url": {"type": "string", "description": "The URL of the article to extract"}
},
},
"summarize_paper": {
"function": tool_summarize_paper,
"description": "Summarizes the content of an academic paper.",
"parameters": {
"paper_text": {"type": "string", "description": "The full text of the paper to be summarized."},
},
},
"reason": {
"function": tool_reason,
"description": "Analyzes and reasons about information.",
"parameters": {
"prompt": {"type": "string", "description": "The original prompt."},
"search_results": {"type": "array", "description": "Search results to analyze."},
"reasoning_context": {"type": "array", "description": "Previous reasoning outputs."},
"critique": {"type": "string", "description": "Recent critique to address."},
"focus_areas": {"type": "array", "description": "Specific aspects to focus on."}
},
},
"summarize": {
"function": tool_summarize,
"description": "Synthesizes insights into a cohesive summary.",
"parameters": {
"insights": {"type": "array", "description": "Insights to summarize."},
"prompt": {"type": "string", "description": "The original research prompt."},
"contradictions": {"type": "array", "description": "Specific contradictions to address."}
},
},
"generate_search_query": {
"function": tool_generate_search_query,
"description": "Generates an optimized search query",
"parameters":{
"prompt": {"type": "string", "description": "The original user prompt."},
"previous_queries": {"type": "array", "description": "Previously used search queries."},
"failed_queries": {"type": "array", "description": "Queries that didn't yield good results."},
"focus_areas": {"type": "array", "description": "Specific aspects to focus on."}
}
},
"critique_reasoning": {
"function": tool_critique_reasoning,
"description": "Critically evaluates reasoning output.",
"parameters": {
"reasoning_output": {"type": "string", "description": "The reasoning output to critique."},
"prompt": {"type": "string", "description": "The original prompt."},
"previous_critiques": {"type": "array", "description": "Previous critique outputs."}
},
},
"identify_contradictions": {
"function": tool_identify_contradictions,
"description": "Identifies contradictions across multiple insights.",
"parameters": {
"insights": {"type": "array", "description": "Collection of insights to analyze for contradictions."},
},
},
"identify_focus_areas": {
"function": tool_identify_focus_areas,
"description": "Identifies specific aspects that need further investigation.",
"parameters": {
"prompt": {"type": "string", "description": "The original research prompt."},
"insights": {"type": "array", "description": "Existing insights to build upon."},
"failed_areas": {"type": "array", "description": "Previously tried areas that yielded poor results."}
},
},
"extract_key_entities": {
"function": tool_extract_key_entities,
"description": "Extracts key entities from the prompt for focused research.",
"parameters": {
"prompt": {"type": "string", "description": "The original research prompt."}
},
},
"meta_analyze": {
"function": tool_meta_analyze,
"description": "Performs meta-analysis across entity-specific insights.",
"parameters": {
"entity_insights": {"type": "object", "description": "Dictionary mapping entities to their insights."},
"prompt": {"type": "string", "description": "The original research prompt."}
},
},
"draft_research_plan": {
"function": tool_draft_research_plan,
"description": "Creates a detailed research plan.",
"parameters": {
"prompt": {"type": "string", "description": "The research question/prompt."},
"entities": {"type": "array", "description": "Key entities to investigate."},
"focus_areas": {"type": "array", "description": "Additional areas to focus on."}
}
},
"search_patents": {
"function": tool_search_patents,
"description": "Searches patent databases globally",
"parameters": {
"query": {"type": "string", "description": "Patent search query"},
"max_results": {"type": "integer", "description": "Maximum number of patents to return"}
}
},
"search_clinical_trials": {
"function": tool_search_clinical_trials,
"description": "Search ClinicalTrials.gov and WHO ICTRP",
"parameters": {
"query": {"type": "string", "description": "Search query for ClinicalTrials.gov and WHO ICTRP"},
"max_results": {"type": "integer", "description": "Maximum number of results to return"}
}
},
"search_datasets": {
"function": tool_search_datasets,
"description": "Search academic datasets from repositories like Kaggle, UCI, etc.",
"parameters": {
"query": {"type": "string", "description": "Search query for academic datasets"},
"max_results": {"type": "integer", "description": "Maximum number of results to return"}
}
},
"search_conferences": {
"function": tool_search_conferences,
"description": "Search major conference proceedings",
"parameters": {
"query": {"type": "string", "description": "Search query for conference proceedings"},
"max_results": {"type": "integer", "description": "Maximum number of results to return"}
}
}
}
}

def create_prompt(task_description, user_input, available_tools, context):
prompt = f"""{task_description}
User Input:
{user_input}
Available Tools:
"""
for tool_name, tool_data in available_tools.items():
prompt += f"- {tool_name}: {tool_data['description']}\n"
prompt += "  Parameters:\n"
for param_name, param_data in tool_data["parameters"].items():
prompt += f"    - {param_name} ({param_data['type']}): {param_data['description']}\n"

recent_context = context[-MAX_CONTEXT_ITEMS:] if len(context) > MAX_CONTEXT_ITEMS else context

prompt += "\nContext (most recent items):\n"
for item in recent_context:
    prompt += f"- {item}\n"

prompt += """
content_copy
download
 Use code with caution.

Instructions:
Select the BEST tool and parameters for the current research stage. Output valid JSON. If no tool is appropriate, respond with {}.
Only use provided tools. Be strategic about which tool to use next based on the research progress so far.
You MUST be methodical.  Think step-by-step:

Plan: If it's the very beginning, extract key entities, identify focus areas, and then draft a research plan.

Search:  Use a variety of search tools.  Start with broad searches, then narrow down.  Use specific search tools (arXiv, PubMed, Scholar) for relevant topics.

Analyze:  Reason deeply about search results, and critique your reasoning.  Identify contradictions. Filter and use FAISS index for relevant information.

Refine: If results are poor, generate better search queries. Adjust focus areas.

Iterate: Repeat steps 2-4, focusing on different entities and aspects.

Synthesize:  Finally, summarize the findings, addressing contradictions.
Example:
{"tool": "search_web", "parameters": {"query": "Eiffel Tower location"}}
Output:
"""
return prompt

def deep_research(prompt):
task_description = "You are an advanced research assistant, designed to be as comprehensive as possible.  Use available tools iteratively, focus on different aspects, explore promising leads thoroughly, critically evaluate your findings, and build up a comprehensive understanding of the research topic. Utilize the FAISS index to avoid redundant searches and to build a persistent knowledge base."
research_data = load_research_data()
paper_summaries = load_paper_summaries()  # Load paper summaries

context = research_data.get('context', [])
all_insights = research_data.get('all_insights', [])
entity_specific_insights = research_data.get('entity_specific_insights', {})
intermediate_output = ""
previous_queries = research_data.get('previous_queries', [])
failed_queries = research_data.get('failed_queries', [])
reasoning_context = research_data.get('reasoning_context', [])
previous_critiques = research_data.get('previous_critiques', [])
focus_areas = research_data.get('focus_areas', [])
failed_areas = research_data.get('failed_areas', [])
seen_snippets = set(research_data.get('seen_snippets', []))
contradictions = research_data.get('contradictions', [])
research_session_id = research_data.get('research_session_id', str(uuid4()))

global index
if research_data:
     logger.info("Restoring FAISS Index from loaded data.")
else:
    index.reset()
    logger.info("Initialized a fresh FAISS Index")

key_entities_with_descriptions = tool_extract_key_entities(prompt=prompt)
key_entities = [e.split(":")[0].strip() for e in key_entities_with_descriptions] # Extract just entity names
if key_entities:
    context.append(f"Identified key entities: {key_entities}")
    intermediate_output += f"Identified key entities for focused research: {key_entities_with_descriptions}\n"
    yield "Identifying key entities... (Completed)"

# Initialize progress tracking for each entity.
entity_progress = {entity: {'queries': [], 'insights': []} for entity in key_entities}
entity_progress['general'] = {'queries': [], 'insights': []}  # For general, non-entity-specific searches
for entity in key_entities + ['general']:
    if entity in research_data:  # Load existing progress
        entity_progress[entity]['queries'] = research_data[entity]['queries']
        entity_progress[entity]['insights'] = research_data[entity]['insights']

if not focus_areas:  # Corrected placement: outside the loop
    initial_focus_areas = tool_identify_focus_areas(prompt=prompt)
    yield "Identifying initial focus areas...(Completed)"
    research_plan = tool_draft_research_plan(prompt=prompt, entities=key_entities, focus_areas=initial_focus_areas)
    yield "Drafting initial research plan...(Completed)"
    context.append(f"Initial Research Plan: {research_plan[:200]}...") # Add plan to context
    intermediate_output += f"Initial Research Plan:\n{research_plan}\n\n"
    focus_areas = initial_focus_areas


for i in range(MAX_ITERATIONS):
    # Entity-focused iteration strategy
    if key_entities and i > 0:  # Cycle through entities *after* initial setup
        entities_to_process = key_entities + ['general']  # Include 'general' for broad searches
        current_entity = entities_to_process[i % len(entities_to_process)]
    else:
        current_entity = 'general'  # Start with general research.

    context.append(f"Current focus: {current_entity}")
    
    # FAISS Retrieval
    if i > 0:  # Use FAISS *after* the first iteration (once we have data)
        faiss_results_indices = search_faiss_index(prompt if current_entity == 'general' else f"{prompt} {current_entity}")
        faiss_context = []
        for idx in faiss_results_indices:
            if idx < len(all_insights):  # Check index bounds
                faiss_context.append(f"Previously found insight: {all_insights[idx]}")
        if faiss_context:
            context.extend(faiss_context) # Add FAISS context
            intermediate_output += f"Iteration {i+1} - Retrieved {len(faiss_context)} relevant items from FAISS index.\n"
            

    if i == 0: #Initial broad search
        initial_query = tool_generate_search_query(prompt=prompt)
        yield f"Generating initial search query... (Iteration {i+1})"
        if initial_query:
            previous_queries.append(initial_query)
            entity_progress['general']['queries'].append(initial_query)

            with ThreadPoolExecutor(max_workers=5) as executor:
                futures = [
                    executor.submit(tool_search_web, query=initial_query, num_results=NUM_RESULTS),
                    executor.submit(tool_search_arxiv, query=initial_query, max_results=5),
                    executor.submit(tool_search_pubmed, query=initial_query, max_results=5),
                    executor.submit(tool_search_wikipedia, query=initial_query, max_results=3),
                    executor.submit(tool_search_scholar, query=initial_query, max_results=5)
                ]

                search_results = []
                for future in as_completed(futures):
                    search_results.extend(future.result())
            yield f"Performing initial searches... (Iteration {i+1})"

            filtered_search_results = filter_results(search_results, prompt)

            if filtered_search_results:
                context.append(f"Initial Search Results: {len(filtered_search_results)} items found")
                reasoning_output = tool_reason(prompt, filtered_search_results)
                yield f"Reasoning about initial search results... (Iteration {i+1})"
                if reasoning_output:
                    all_insights.append(reasoning_output)
                    entity_progress['general']['insights'].append(reasoning_output)
                    reasoning_context.append(reasoning_output)
                    context.append(f"Initial Reasoning: {reasoning_output[:200]}...")
                    add_to_faiss_index(reasoning_output)
            else:
                failed_queries.append(initial_query)
                context.append(f"Initial query yielded no relevant results: {initial_query}")

    elif current_entity != 'general':
        entity_query = tool_generate_search_query(
            prompt=f"{prompt} focusing specifically on {current_entity}",
            previous_queries=entity_progress[current_entity]['queries'],
            focus_areas=focus_areas
        )
        yield f"Generating search query for entity: {current_entity}... (Iteration {i+1})"

        if entity_query:
            previous_queries.append(entity_query)
            entity_progress[current_entity]['queries'].append(entity_query)


            with ThreadPoolExecutor(max_workers=5) as executor:
                futures = [
                    executor.submit(tool_search_web, query=entity_query, num_results=NUM_RESULTS//2),
                    executor.submit(tool_search_arxiv, query=entity_query, max_results=3),
                    executor.submit(tool_search_pubmed, query=entity_query, max_results=3),
                    executor.submit(tool_search_wikipedia, query=entity_query, max_results=2),
                    executor.submit(tool_search_scholar, query=entity_query, max_results=3)
                ]

                search_results = []
                for future in as_completed(futures):
                    search_results.extend(future.result())

            yield f"Searching for information on entity: {current_entity}... (Iteration {i+1})"
            filtered_search_results = filter_results(search_results,
                                                    f"{prompt} {current_entity}",
                                                    previous_snippets=seen_snippets) # Pass existing snippets

            if filtered_search_results:
                context.append(f"Entity Search for {current_entity}: {len(filtered_search_results)} results")

                entity_reasoning = tool_reason(
                    prompt=f"{prompt} focusing on {current_entity}",
                    search_results=filtered_search_results,
                    reasoning_context=entity_progress[current_entity]['insights'], # Use entity-specific context
                    focus_areas=focus_areas
                )
                yield f"Reasoning about entity: {current_entity}... (Iteration {i+1})"

                if entity_reasoning:
                    all_insights.append(entity_reasoning)
                    entity_progress[current_entity]['insights'].append(entity_reasoning)

                    if current_entity not in entity_specific_insights:
                        entity_specific_insights[current_entity] = []
                    entity_specific_insights[current_entity].append(entity_reasoning)

                    context.append(f"Reasoning about {current_entity}: {entity_reasoning[:200]}...")
                    add_to_faiss_index(entity_reasoning)
            else:
                failed_queries.append(entity_query)
                context.append(f"Entity query for {current_entity} yielded no relevant results")

    llm_prompt = create_prompt(task_description, prompt, tools, context)
    llm_response = hf_inference(MAIN_LLM_MODEL, llm_prompt, stream=True)  # Use streaming

    if isinstance(llm_response, dict) and "error" in llm_response:
        intermediate_output += f"LLM Error: {llm_response['error']}\n"
        yield f"LLM Error (Iteration {i+1}): {llm_response['error']}"  # Display error in output
        continue

    # Process streaming response
    response_text = ""
    try:
        for chunk in llm_response:
            if chunk.choices and chunk.choices[0].delta and chunk.choices[0].delta.content:
                response_text += chunk.choices[0].delta.content
                yield f"Iteration {i+1} - Thinking... {response_text}" # Real time output

    except Exception as e:
        intermediate_output += f"Streaming Error: {str(e)}\n"
        yield f"Streaming Error (Iteration {i+1}): {str(e)}" #Error
        continue

    try:
        response_json = json.loads(response_text)  # Parse the JSON response.
        intermediate_output += f"Iteration {i+1} - Focus: {current_entity} - Action: {response_text}\n"
    except json.JSONDecodeError:
        intermediate_output += f"Iteration {i+1} - LLM Response (Invalid JSON): {response_text[:100]}...\n"
        context.append(f"Invalid JSON: {response_text[:100]}...") # Add invalid JSON to context
        continue

    tool_name = response_json.get("tool")
    parameters = response_json.get("parameters", {})

    if not tool_name: #LLM didn't return a tool.  End the process if we are past halfway.
        if all_insights:
            if i > MAX_ITERATIONS // 2:
                break
        continue

    if tool_name not in tools:
        context.append(f"Invalid tool: {tool_name}")
        intermediate_output += f"Iteration {i + 1} - Invalid tool chosen: {tool_name}\n"
        continue

    tool = tools[tool_name]
    try:
        intermediate_output += f"Iteration {i+1} - Executing: {tool_name}, Key params: {str(parameters)[:100]}...\n"

        if tool_name == "generate_search_query":
            parameters['previous_queries'] = previous_queries
            parameters['failed_queries'] = failed_queries
            parameters['focus_areas'] = focus_areas
            result = tool["function"](**parameters)
            yield f"Iteration {i+1} - Generated search query: {result}"

            if current_entity != 'general':
                entity_progress[current_entity]['queries'].append(result) # Add entity-specific

            previous_queries.append(result)

        elif tool_name in ["search_web", "search_arxiv", "search_pubmed", "search_wikipedia", "search_scholar"]:
            result = tool["function"](**parameters)
            search_prompt = prompt
            if current_entity != 'general':
                search_prompt = f"{prompt} focusing on {current_entity}"

            filtered_result = filter_results(result, search_prompt, previous_snippets=seen_snippets)

            result = filtered_result  # Work with filtered results

            if not result and 'query' in parameters: # Add query to failures if nothing returned.
                failed_queries.append(parameters['query'])

        elif tool_name == "reason":
            # Ensure correct reasoning context is passed.
            if current_entity != 'general' and 'reasoning_context' not in parameters:
                parameters['reasoning_context'] = entity_progress[current_entity]['insights']
            elif 'reasoning_context' not in parameters:
                parameters['reasoning_context'] = reasoning_context[:]

            if 'prompt' not in parameters:
                if current_entity != 'general':
                    parameters['prompt'] = f"{prompt} focusing on {current_entity}"
                else:
                    parameters['prompt'] = prompt

            if 'search_results' not in parameters:
                parameters['search_results'] = [] #Avoid errors if no search results.

            if 'focus_areas' not in parameters and focus_areas: # Avoid overwriting focus_areas if already set
                parameters['focus_areas'] = focus_areas

            result = tool["function"](**parameters)
            yield f"Iteration {i+1} - Reasoning about information..."

            if current_entity != 'general':
                entity_progress[current_entity]['insights'].append(result)
                if current_entity not in entity_specific_insights:
                     entity_specific_insights[current_entity] = []
                entity_specific_insights[current_entity].append(result)
            else:
                reasoning_context.append(result) #Add to general context.
            add_to_faiss_index(result)
            all_insights.append(result)

        elif tool_name == "critique_reasoning":
            if 'previous_critiques' not in parameters: #Pass in the previous critiques.
                parameters['previous_critiques'] = previous_critiques

            if all_insights:
                if 'reasoning_output' not in parameters:
                    parameters['reasoning_output'] = all_insights[-1]  #Critique the most recent insight.
                if 'prompt' not in parameters:
                    parameters['prompt'] = prompt

                result = tool["function"](**parameters)
                yield f"Iteration {i+1} - Critiquing reasoning..."
                previous_critiques.append(result)
                context.append(f"Critique: {result[:200]}...")
            else:
                result = "No reasoning to critique yet."

        elif tool_name == "identify_contradictions":
            result = tool["function"](**parameters)
            yield f"Iteration {i+1} - Identifying contradictions..."
            if result:
                contradictions = result  # Keep track of contradictions.
                context.append(f"Identified contradictions: {result}")

        elif tool_name == "identify_focus_areas":
            if 'failed_areas' not in parameters:
                parameters['failed_areas'] = failed_areas
            result = tool["function"](**parameters)
            yield f"Iteration {i+1} - Identifying focus areas..."
            if result:
                old_focus = set(focus_areas)
                focus_areas = result  # Update focus areas
                failed_areas.extend([area for area in old_focus if area not in result])  #Track failed areas
                context.append(f"New focus areas: {result}")

        elif tool_name == "extract_article":
            result = tool["function"](**parameters)
            yield f"Iteration {i+1} - Extracting article content..."
            if result:
                context.append(f"Extracted article content from {parameters['url']}: {result[:200]}...")
                # Reason specifically about the extracted article.
                reasoning_about_article = tool_reason(prompt=prompt, search_results=[{"title": "Extracted Article", "snippet": result, "url": parameters['url']}])
                if reasoning_about_article:
                    all_insights.append(reasoning_about_article)
                    add_to_faiss_index(reasoning_about_article)
        
        elif tool_name == "summarize_paper":
            result = tool["function"](**parameters)
            yield f"Iteration {i+1} - Summarizing paper..."
            if result:
                paper_summaries[parameters['paper_text'][:100]] = result  # Store by a snippet of the text
                save_paper_summaries(paper_summaries)
                context.append(f"Summarized paper: {result[:200]}...")
                add_to_faiss_index(result) # Add the summary itself to FAISS.
                all_insights.append(result) #Add summary to insights for later summarization.

        elif tool_name == "meta_analyze":
            if 'entity_insights' not in parameters:
                parameters['entity_insights'] = entity_specific_insights
            if 'prompt' not in parameters:
                parameters['prompt'] = prompt
            result = tool["function"](**parameters)
            yield f"Iteration {i+1} - Performing meta-analysis..."
            if result:
                all_insights.append(result)  # Add meta-analysis to overall insights.
                context.append(f"Meta-analysis across entities: {result[:200]}...")
                add_to_faiss_index(result)


        elif tool_name == "draft_research_plan":
            result = "Research plan already generated."  # Avoid re-generating.

        else:
            result = tool["function"](**parameters)

        result_str = str(result)
        if len(result_str) > 500:
            result_str = result_str[:500] + "..."

        intermediate_output += f"Iteration {i+1} - Result: {result_str}\n"

        # Add tool use to context, limit context length
        result_context = result_str
        if len(result_str) > 300:
            result_context = result_str[:300] + "..."
        context.append(f"Used: {tool_name}, Result: {result_context}")

    except Exception as e:
        logger.error(f"Error with {tool_name}: {str(e)}")
        context.append(f"Error with {tool_name}: {str(e)}")
        intermediate_output += f"Iteration {i+1} - Error: {str(e)}\n"
        continue

    #Save data
    research_data = {
        'context': context,
        'all_insights': all_insights,
        'entity_specific_insights': entity_specific_insights,
        'previous_queries': previous_queries,
        'failed_queries': failed_queries,
        'reasoning_context': reasoning_context,
        'previous_critiques': previous_critiques,
        'focus_areas': focus_areas,
        'failed_areas': failed_areas,
        'seen_snippets': list(seen_snippets),
        'contradictions': contradictions,
        'research_session_id': research_session_id
    }
    for entity in entity_progress:
         research_data[entity] = entity_progress[entity] #save the individual entity
    save_research_data(research_data, index)


# Perform meta-analysis *before* final summarization, if we have enough entity-specific insights.
if len(entity_specific_insights) > 1 and len(all_insights) > 2:
    meta_analysis = tool_meta_analyze(entity_insights=entity_specific_insights, prompt=prompt)
    if meta_analysis:
        all_insights.append(meta_analysis)
        intermediate_output += f"Final Meta-Analysis: {meta_analysis[:500]}...\n"
        add_to_faiss_index(meta_analysis)  # Add to FAISS

if all_insights:
    final_result = tool_summarize(all_insights, prompt, contradictions) # Summarize all insights.
else:
    final_result = "Could not find meaningful information despite multiple attempts."


full_output = f"**Research Prompt:** {prompt}\n\n"

if key_entities_with_descriptions:
     full_output += f"**Key Entities Identified:**\n"
     for entity in key_entities_with_descriptions:
         full_output += f"- {entity}\n"
     full_output += "\n"

full_output += "**Research Process:**\n" + intermediate_output + "\n"

if contradictions:
    full_output += "**Contradictions Identified:**\n"
    for i, contradiction in enumerate(contradictions, 1):
        full_output += f"{i}. {contradiction}\n"
    full_output += "\n"

full_output += f"**Final Analysis:**\n{final_result}\n\n"

full_output += f"Research Session ID: {research_session_id}\n"
full_output += f"Completed at: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}\n"
full_output += f"Total iterations: {i+1}\n"
full_output += f"Total insights generated: {len(all_insights)}\n"

yield full_output  # Final output
content_copy
download
 Use code with caution.

custom_css = """
/* Modern Research Interface */
.research-container {
    display: grid;
    grid-template-columns: 2fr 1fr;
    gap: 20px;
    padding: 20px;
    background: #1a1a1a;
}

.main-content {
    background: #2d2d2d;
    border-radius: 10px;
    padding: 20px;
}

.sidebar {
    background: #2d2d2d;
    border-radius: 10px;
    padding: 20px;
}

/* Progress Tracking */
.progress-container {
    background: #333;
    border-radius: 8px;
    padding: 15px;
    margin-bottom: 15px;
}

.tool-indicator {
    display: flex;
    align-items: center;
    gap: 10px;
    padding: 8px;
    background: #444;
    border-radius: 5px;
    margin-bottom: 8px;
}

.tool-icon {
    width: 24px;
    height: 24px;
}

.tool-name {
    color: #4CAF50;
    font-weight: bold;
}

/* Enhanced Output Formatting */
.research-output {
    font-family: 'Inter', sans-serif;
    line-height: 1.6;
}

.research-output h2 {
    color: #4CAF50;
    border-bottom: 2px solid #4CAF50;
    padding-bottom: 5px;
}

.research-output code {
    background: #333;
    padding: 2px 6px;
    border-radius: 4px;
}

/* Statistics Panel */
.stats-panel {
    background: #333;
    border-radius: 8px;
    padding: 15px;
    margin-top: 15px;
}

.stat-item {
    display: flex;
    justify-content: space-between;
    padding: 8px 0;
    border-bottom: 1px solid #444;
}
"""

iface = gr.Interface(
    fn=deep_research,
    inputs=[
        gr.Textbox(lines=5, placeholder="Enter your research question...", label="Research Question"),
        gr.Dropdown(
            choices=["Quick", "Standard", "Comprehensive", "Exhaustive"],
            label="Research Depth",
            value="Standard"
        ),
        gr.CheckboxGroup(
            choices=["Academic Papers", "Patents", "News", "Clinical Trials", "Datasets"],
            label="Source Types",
            value=["Academic Papers", "News"]
        ),
        gr.Slider(
            minimum=1,
            maximum=24,
            value=2,
            label="Time Limit (hours)"
        )
    ],
    outputs=[
        gr.Markdown(label="Research Results", elem_classes=["research-output"]),
        gr.JSON(label="Progress Statistics", elem_classes=["stats-panel"]),
        gr.HTML(label="Active Tools", elem_classes=["tool-indicator"])
    ],
    title="Advanced Research Institution Assistant",
    description="""Enterprise-grade research system with real-time progress tracking, 
    comprehensive source coverage, and advanced analysis capabilities.""",
    theme=gr.themes.Base(primary_hue="green"),
    css=custom_css
)

if name == "main":
iface.launch(share=False)