Readme update and cleanup

README.md

@@ -9,28 +9,29 @@ app_file: app.py
 fullWidth: true
 pinned: false
 startup_duration_timeout: 1h
+license: apache-2.0
+short_description: AI-powered conversational assistant for EU Deforestation Regulation compliance
 models:
 - BAAI/bge-m3
 - BAAI/bge-reranker-v2-m3
 - meta-llama/Llama-3.1-8B-Instruct
 ---
-short_description: AI-powered conversational assistant for EU Deforestation Regulation compliance
-license: apache-2.0
+
 
 
 ## Technical Documentation of the system in accordance with EU AI Act
 
-**System Name:** EUDR ChatBot
+**System Name:** EUDR Chatbot
 
 **Provider / Supplier:** GIZ Data Service Center
 
-**As of:** July 2025
+**As of:** September 2025
 
 ## 1. General Description of the System
 
 EUDR Bot is an AI-powered conversational assistant designed to help you understand compliance with and analyze the EU Deforestation Regulation. This tool leverages advanced language models to help you get clear and structured answers about EUDR requirements, compliance procedures, and regulatory guidance.
 
-It combines a generative language assistant with a knowledge base implemented via Retrieval-Augmented Generation (RAG). The scope and functionality of the tool is focused on EU Deforestation Regulation compliance and related documentation.
+It combines a generative language assistant with a knowledge base implemented via Retrieval-Augmented Generation (RAG). In addition to the RAG, the tool also provide quick analysis on the geojson of plot by leveraging the [Whisp](https://openforis.org/solutions/whisp/). The scope and functionality of the tool is focused on EU Deforestation Regulation compliance and related documentation in context of 'Ecuador' and 'Guatemala' only.
 
 ## 2. Models Used
 
@@ -46,6 +47,10 @@ It combines a generative language assistant with a knowledge base implemented vi
 - **Model Name:** [BAAI/bge-reranker-v2-m3](https://huggingface.co/BAAI/bge-reranker-v2-m3)
 - **Model Source:** Local Instance
 
+### External API's
+- **Whisp Info**: https://openforis.org/solutions/whisp/
+- **API Endpoint** https://whisp.openforis.org/documentation/api-guide
+
 ## 3. Model Training Data
 
 All the models mentioned above are being consumed without any fine-tuning or training being performed by the developer team of EUDR Bot. And hence there is no training data which had been used by the development team of EUDR Bot.
@@ -57,7 +62,7 @@ All the models mentioned above are being consumed without any fine-tuning or tra
 - **Embedding Dimension:** 1024
 - **Vector Database:** Qdrant (via API)
 - **Framework:** Langchain (custom RAG pipeline)
-- **Top-k:** 5 relevant text segments per query
+- **Top-k:** 10 relevant text segments per query
 
 ## 5. System Limitations and Non-Purposes
 
@@ -71,14 +76,14 @@ All the models mentioned above are being consumed without any fine-tuning or tra
 - The user interface clearly indicates the use of a generative AI model.
 - An explanation of the RAG method is included.
 - We collect usage statistics as detailed in Disclaimer tab of the app along with the explicit display in the user interface of the tool.
-- Feedback mechanism available (via https://huggingface.co/spaces/GIZ/eudr_assistant/discussions/new).
+- Feedback mechanism available (via https://huggingface.co/spaces/GIZ/Asistente_EUDR/discussions/new).
 
 ## 7. Monitoring, Feedback, and Incident Reporting
 
-- User can provide feedback via UI by giving (Thumbs-up or down to AI-Generated answer). Alternatively for more detailed feedback please use https://huggingface.co/spaces/GIZ/eudr_assistant/discussions/new to report any issue.
+- User can provide feedback via UI by giving (Thumbs-up or down to AI-Generated answer). Alternatively for more detailed feedback please use https://huggingface.co/spaces/GIZ/Asistente_EUDR/discussions/new to report any issue.
 - Technical development is carried out by the GIZ Data Service Center.
 - No automated bias detection – but low risk due to content restrictions.
 
 ## 8. Contact
 
-For any questions, please contact via https://huggingface.co/spaces/GIZ/eudr_assistant/discussions/new or send us email to [email protected]
+For any questions, please contact via https://huggingface.co/spaces/GIZ/Asistente_EUDR/discussions/new or send us email to [email protected]

app1.py

@@ -1,763 +0,0 @@
-import gradio as gr
-import pandas as pd
-import logging
-import asyncio
-import os
-import time
-from uuid import uuid4
-from datetime import datetime, timedelta
-from pathlib import Path
-from huggingface_hub import CommitScheduler, HfApi
-from auditqa.sample_questions import QUESTIONS
-from auditqa.reports import files, report_list, new_files, new_report_list
-from auditqa.process_chunks import load_chunks, getconfig, get_local_qdrant
-from auditqa.retriever import get_context
-from auditqa.reader import nvidia_client, dedicated_endpoint, serverless_api, inf_provider
-from auditqa.utils import make_html_source, parse_output_llm_with_sources, save_logs, get_message_template, get_client_location, get_client_ip, get_platform_info
-from dotenv import load_dotenv
-load_dotenv()
-from threading import Lock
-from gradio.routes import Request
-import json
-#import platform
-#print(platform.python_version())
-
-# fetch tokens and model config params
-SPACES_LOG = os.environ["SPACES_LOG"]
-#audit_space = os.environ["audit_space"]
-model_config = getconfig("model_params.cfg")
-
-# create the local logs repo
-JSON_DATASET_DIR = Path("json_dataset")
-JSON_DATASET_DIR.mkdir(parents=True, exist_ok=True)
-JSON_DATASET_PATH = JSON_DATASET_DIR / f"logs-{uuid4()}.json"
-
-# the logs are written to dataset repo periodically from local logs
-# https://huggingface.co/spaces/Wauplin/space_to_dataset_saver
-scheduler = CommitScheduler(
-     repo_id="GIZ/spaces_logs",
-     repo_type="dataset",
-     folder_path=JSON_DATASET_DIR,
-     path_in_repo="audit_chatbot",
-     token=SPACES_LOG )
-
-#####--------------- VECTOR STORE -------------------------------------------------
-# reports contain the already created chunks from Markdown version of pdf reports
-# document processing was done using : https://github.com/axa-group/Parsr
-# We need to create the local vectorstore collection once using load_chunks
-# vectorestore colection are stored on persistent storage so this needs to be run only once
-# hence, comment out line below when creating for first time
-#vectorstores = load_chunks()
-# once the vectore embeddings  are created we will use qdrant client to access these
-#try:
-vectorstores = get_local_qdrant()
-#except Exception as e:
-#    api = HfApi()
-#    api.restart_space(repo_id = "GIZ/audit_assistant", token=audit_space)
-
-#####---------------------CHAT-----------------------------------------------------
-def start_chat(query,history):
-    history = history + [(query,None)]
-    history = [tuple(x) for x in history]
-    return (gr.update(interactive = False),gr.update(selected=1),history)
-
-def finish_chat():
-    return (gr.update(interactive = True,value = ""))
-
-def submit_feedback(feedback, logs_data):
-    """Handle feedback submission"""
-    try:
-        if logs_data is None:
-            return gr.update(visible=False), gr.update(visible=True)
-            
-        session_id = logs_data.get("session_id")
-        if session_id:
-            # Update session last_activity to now
-            session_manager.update_session(session_id)
-            # Compute duration from the session manager and update the log.
-            logs_data["session_duration_seconds"] = session_manager.get_session_duration(session_id)
-            
-        # Now save the (feedback) log record
-        save_logs(scheduler, JSON_DATASET_PATH, logs_data, feedback)
-        return gr.update(visible=False), gr.update(visible=True)
-    except Exception as e:
-        return gr.update(visible=False), gr.update(visible=True)
-
-
-# Session Manager added (track session duration, location, and platform)
-class SessionManager:
-    def __init__(self):
-        self.sessions = {}
-        
-    def create_session(self, client_ip, user_agent):
-        session_id = str(uuid4())
-        self.sessions[session_id] = {
-            'start_time': datetime.now(),
-            'last_activity': datetime.now(),
-            'client_ip': client_ip,
-            'location_info': get_client_location(client_ip),
-            'platform_info': get_platform_info(user_agent)
-        }
-        return session_id
-    
-    def update_session(self, session_id):
-        if session_id in self.sessions:
-            self.sessions[session_id]['last_activity'] = datetime.now()
-    
-    def get_session_duration(self, session_id):
-        if session_id in self.sessions:
-            start = self.sessions[session_id]['start_time']
-            last = self.sessions[session_id]['last_activity']
-            return (last - start).total_seconds()
-        return 0
-    
-    def get_session_data(self, session_id):
-        return self.sessions.get(session_id)
-
-# Initialize session manager
-session_manager = SessionManager()
-    
-async def chat(query,history, method, sources,reports,subtype, client_ip=None, session_id = None, request:gr.Request = None):
-    """taking a query and a message history, use a pipeline (reformulation, retriever, answering) 
-       to yield a tuple of:(messages in gradio format/messages in langchain format, source documents)
-    """
-
-    if not session_id:
-        user_agent =  request.headers.get('User-Agent','') if request else ''
-        session_id = session_manager.create_session(client_ip, user_agent)
-    else:
-        session_manager.update_session(session_id)
-
-    # Get session id
-    session_data = session_manager.get_session_data(session_id)
-    session_duration = session_manager.get_session_duration(session_id)
-
-    print(f">> NEW QUESTION : {query}")
-    print(f"history:{history}")
-    print(f"sources:{sources}")
-    print(f"reports:{reports}")
-    print(f"subtype:{subtype}")
-    #print(f"year:{year}")
-    docs_html = ""
-    output_query = ""
-    if method == "Search by Report Name":
-        if len(reports) == 0:
-            warning_message = "⚠️ **No Report Selected.** Please select the relevant reports."
-            history[-1] = (query, warning_message)
-            # Update logs with the warning instead of answer
-            logs_data = {
-                "record_id": str(uuid4()),
-                "session_id": session_id,
-                "session_duration_seconds": session_duration,
-                "client_location": session_data['location_info'],
-                "platform": session_data['platform_info'],
-                "question": query,
-                "retriever": model_config.get('retriever','MODEL'),
-                "endpoint_type": model_config.get('reader','TYPE'),
-                "reader": model_config.get('reader','NVIDIA_MODEL'),
-                "answer": warning_message,
-                "no_results": True  # Flag to indicate no results were found
-            }
-            yield [tuple(x) for x in history], "", logs_data, session_id
-            # Save log for the warning response
-            save_logs(scheduler, JSON_DATASET_PATH, logs_data)
-            return
-    else:
-        if sources is None and subtype is None:
-            warning_message = "⚠️ **No Report Selected.** Please select the relevant reports."
-            history[-1] = (query, warning_message)
-            # Update logs with the warning instead of answer
-            logs_data = {
-                "record_id": str(uuid4()),
-                "session_id": session_id,
-                "session_duration_seconds": session_duration,
-                "client_location": session_data['location_info'],
-                "platform": session_data['platform_info'],
-                "question": query,
-                "retriever": model_config.get('retriever','MODEL'),
-                "endpoint_type": model_config.get('reader','TYPE'),
-                "reader": model_config.get('reader','NVIDIA_MODEL'),
-                "answer": warning_message,
-                "no_results": True  # Flag to indicate no results were found
-            }
-            yield [tuple(x) for x in history], "", logs_data, session_id
-            # Save log for the warning response
-            save_logs(scheduler, JSON_DATASET_PATH, logs_data)
-            return
-            
-
-    ##------------------------fetch collection from vectorstore------------------------------
-    vectorstore = vectorstores["docling"]
-
-    ##------------------------------get context---------------------------------------------- 
-
-    ### adding for assessing computation time
-    start_time = time.time()
-    context_retrieved = get_context(vectorstore=vectorstore,query=query,reports=reports,
-                                                sources=sources,subtype=subtype)
-    end_time = time.time()
-    print("Time for retriever:",end_time - start_time)
-
-        
-    if not context_retrieved or len(context_retrieved) == 0:
-        warning_message = "⚠️ **No relevant information was found in the audit reports pertaining your query.** Please try rephrasing your question or selecting different report filters."
-        history[-1] = (query, warning_message)
-        # Update logs with the warning instead of answer
-        logs_data = {
-            "record_id": str(uuid4()),
-            "session_id": session_id,
-            "session_duration_seconds": session_duration,
-            "client_location": session_data['location_info'],
-            "platform": session_data['platform_info'],
-            "question": query,
-            "retriever": model_config.get('retriever','MODEL'),
-            "endpoint_type": model_config.get('reader','TYPE'),
-            "reader": model_config.get('reader','NVIDIA_MODEL'),
-            "answer": warning_message,
-            "no_results": True  # Flag to indicate no results were found
-        }
-        yield [tuple(x) for x in history], "", logs_data, session_id
-        # Save log for the warning response
-        save_logs(scheduler, JSON_DATASET_PATH, logs_data)
-        return
-    context_retrieved_formatted = "||".join(doc.page_content for doc in context_retrieved)
-    context_retrieved_lst = [doc.page_content for doc in context_retrieved]
-    
-    ##------------------- -------------Define Prompt-------------------------------------------
-    SYSTEM_PROMPT = """
-        You are AuditQ&A, an AI Assistant created by Auditors and Data Scientist. \
-            You are given a question and extracted passages of the consolidated/departmental/thematic focus audit reports.\
-            Provide a clear and structured answer based on the passages/context provided and the guidelines.
-        Guidelines:
-        - Passeges are provided as comma separated list of strings
-        - If the passages have useful facts or numbers, use them in your answer.
-        - When you use information from a passage, mention where it came from by using [Doc i] at the end of the sentence. i stands for the number of the document.
-        - Do not use the sentence 'Doc i says ...' to say where information came from.
-        - If the same thing is said in more than one document, you can mention all of them like this: [Doc i, Doc j, Doc k]
-        - Do not just summarize each passage one by one. Group your summaries to highlight the key parts in the explanation.
-        - If it makes sense, use bullet points and lists to make your answers easier to understand.
-        - You do not need to use every passage. Only use the ones that help answer the question.
-        - If the documents do not have the information needed to answer the question, just say you do not have enough information.
-        """
-    
-    USER_PROMPT = """Passages:
-        {context}
-        -----------------------
-        Question: {question}  - Explained to audit expert
-        Answer in english with the passages citations:
-        """.format(context = context_retrieved_lst, question=query)
-    
-    ##-------------------- apply message template ------------------------------
-    messages = get_message_template(model_config.get('reader','TYPE'),SYSTEM_PROMPT,USER_PROMPT)
-
-    ## -----------------Prepare HTML for displaying source documents --------------
-    docs_html = []
-    for i, d in enumerate(context_retrieved, 1):
-        docs_html.append(make_html_source(d, i))
-    docs_html = "".join(docs_html)
-
-    ##-----------------------get answer from endpoints------------------------------
-    answer_yet = ""
-
-    logs_data = {
-        "record_id": str(uuid4()),  # Add unique record ID
-        "session_id": session_id,
-        "session_duration_seconds": session_duration,
-        "client_location": session_data['location_info'],
-        "platform": session_data['platform_info'], 
-        "system_prompt": SYSTEM_PROMPT, 
-        "sources": sources, 
-        "reports": reports, 
-        "subtype": subtype, 
-        #"year": year,
-        "question": query,
-        "retriever": model_config.get('retriever','MODEL'),
-        "endpoint_type": model_config.get('reader','TYPE'),
-        "reader": model_config.get('reader','NVIDIA_MODEL'),
-        "docs": [doc.page_content for doc in context_retrieved],
-    }
-
-
-    if model_config.get('reader','TYPE') == 'NVIDIA':
-        chat_model = nvidia_client()
-        async def process_stream():
-            nonlocal answer_yet # Use the outer scope's answer_yet variable
-            # Without nonlocal, Python would create a new local variable answer_yet inside process_stream(), 
-            # instead of modifying the one from the outer scope.
-            # Iterate over the streaming response chunks
-            response = chat_model.chat_completion(
-                                model=model_config.get("reader","NVIDIA_MODEL"),
-                                messages=messages,
-                                stream=True,
-                                max_tokens=int(model_config.get('reader','MAX_TOKENS')),
-                            )
-            for message in response:
-                token = message.choices[0].delta.content
-                if token:
-                    answer_yet += token
-                    parsed_answer = parse_output_llm_with_sources(answer_yet)
-                    history[-1] = (query, parsed_answer)
-                    logs_data["answer"] = parsed_answer
-                    yield [tuple(x) for x in history], docs_html, logs_data, session_id
-
-        # Stream the response updates
-        async for update in process_stream():
-            yield update
-
-    elif model_config.get('reader','TYPE') == 'INF_PROVIDERS':
-        chat_model = inf_provider()
-        start_time = time.time()
-        ai_prefix = "**AI-Generated Response:**\n\n"
-        async def process_stream():
-            nonlocal answer_yet # Use the outer scope's answer_yet variable
-            # Without nonlocal, Python would create a new local variable answer_yet inside process_stream(), 
-            # instead of modifying the one from the outer scope.
-            # Iterate over the streaming response chunks
-            answer_yet += ai_prefix
-            response = chat_model.chat.completions.create(
-                                        model=model_config.get("reader","INF_PROVIDER_MODEL"),
-                                        messages = messages,
-                                        stream= True,
-                                        max_tokens=int(model_config.get('reader','MAX_TOKENS')),              
-                                    )
-            for message in response:
-                token = message.choices[0].delta.content
-                if token:
-                    answer_yet += token
-                    parsed_answer = parse_output_llm_with_sources(answer_yet)
-                    history[-1] = (query, parsed_answer)
-                    logs_data["answer"] = parsed_answer
-                    yield [tuple(x) for x in history], docs_html, logs_data, session_id
-                    await asyncio.sleep(0.05)
-
-        # Stream the response updates
-        async for update in process_stream():
-            yield update
-            
-
-    elif model_config.get('reader','TYPE') == 'DEDICATED':
-        chat_model = dedicated_endpoint()
-        ### adding for assessing computation time
-        start_time = time.time()
-        async def process_stream():
-        # Without nonlocal, Python would create a new local variable answer_yet inside process_stream(), 
-        # instead of modifying the one from the outer scope.
-            nonlocal answer_yet # Use the outer scope's answer_yet variable
-            # Iterate over the streaming response chunks
-            async for chunk in chat_model.astream(messages):
-                token = chunk.content
-                answer_yet += token
-                parsed_answer = parse_output_llm_with_sources(answer_yet)
-                history[-1] = (query, parsed_answer)
-                logs_data["answer"] = parsed_answer
-                yield [tuple(x) for x in history], docs_html, logs_data, session_id
-        end_time = time.time()
-        print("Time for reader:",end_time - start_time)
-        
-       # Stream the response updates
-        async for update in process_stream():
-            yield update
-        
-
-    else:
-        chat_model = serverless_api() # TESTING: ADAPTED FOR HF INFERENCE API (needs to be reverted for production version)
-        async def process_stream():
-            nonlocal answer_yet
-            try:
-                formatted_messages = [
-                    {
-                        "role": msg.type if hasattr(msg, 'type') else msg.role,
-                        "content": msg.content
-                    }
-                    for msg in messages
-                ]
-                
-                response = chat_model.chat_completion(
-                    messages=formatted_messages,
-                    max_tokens= int(model_config.get('reader', 'MAX_TOKENS'))
-                )
-                
-                response_text = response.choices[0].message.content
-                words = response_text.split()
-                for word in words:
-                    answer_yet += word + " "
-                    parsed_answer = parse_output_llm_with_sources(answer_yet)
-                    history[-1] = (query, parsed_answer)
-                    # Update logs_data with current answer (and get a new timestamp)
-                    logs_data["answer"] = parsed_answer
-                    yield [tuple(x) for x in history], docs_html, logs_data, session_id
-                    await asyncio.sleep(0.05)
-                
-            except Exception as e:
-                raise
-
-        async for update in process_stream():
-            yield update
-    
-
-    # logging the event
-    try:
-        save_logs(scheduler,JSON_DATASET_PATH,logs_data)
-    except Exception as e:
-        raise
-
-
-
-
-#####-------------------------- Gradio App--------------------------------------####
-
-# Set up Gradio Theme
-theme = gr.themes.Base(
-    primary_hue="blue",
-    secondary_hue="red",
-    font=[gr.themes.GoogleFont("Poppins"), "ui-sans-serif", "system-ui", "sans-serif"],
-    text_size = gr.themes.utils.sizes.text_sm,
-)
-
-init_prompt =  """
-Hello, I am Audit Q&A, an AI-powered conversational assistant designed to help you understand audit reports. I will answer your questions by using **Audit reports publishsed by Auditor General Office, Uganda**.
-💡 How to use (tabs on right)
-- **Reports**: You can choose to address your question to either specific report or a collection of report like District or Ministry focused reports. \
-If you dont select any then the Consolidated report is relied upon to answer your question.
-- **Examples**: We have curated some example questions,select a particular question from category of questions.
-- **Sources**: This tab will display the relied upon paragraphs from the report, to help you in assessing or fact checking if the answer provided by Audit Q&A assitant is correct or not.
-⚠️ For limitations of the tool and collection of usage statistics and data please check **Disclaimer** tab.
-⚠️ By using this app, you acknowledge that we collect usage statistics (such as question asked, feedback given, session duration, device type and anonymized geo-location) to understand performance and continuously improve the tool, based on our legitimate interest in enhancing our services.
-"""
-
-
-with gr.Blocks(title="Audit Q&A", css= "style.css", theme=theme,elem_id = "main-component") as demo:
-    #----------------------------------------------------------------------------------------------
-    # main tab where chat interaction happens
-    # ---------------------------------------------------------------------------------------------
-    with gr.Tab("AuditQ&A"):
-        
-        with gr.Row(elem_id="chatbot-row"):
-            # chatbot output screen
-            with gr.Column(scale=2):
-                chatbot = gr.Chatbot(
-                    value=[(None,init_prompt)],
-                    show_copy_button=True,show_label = False,elem_id="chatbot",layout = "panel",
-                    avatar_images = (None,"data-collection.png"),
-                )
-
-                # feedback UI
-                with gr.Column(elem_id="feedback-container"):
-                    with gr.Row(visible=False) as feedback_row:
-                        gr.Markdown("Was this response helpful?")
-                        with gr.Row():
-                            okay_btn = gr.Button("👍 Okay", elem_classes="feedback-button")
-                            not_okay_btn = gr.Button("👎 Not to expectations", elem_classes="feedback-button")
-                    feedback_thanks = gr.Markdown("Thanks for the feedback!", visible=False)
-                    feedback_state = gr.State()
-                
-
-
-
-                with gr.Row(elem_id = "input-message"):
-                    textbox=gr.Textbox(placeholder="Ask me anything here!",show_label=False,scale=7,
-                                       lines = 1,interactive = True,elem_id="input-textbox")
-
-            # second column with playground area for user to select values
-            with gr.Column(scale=1, variant="panel",elem_id = "right-panel"):
-                # creating tabs on right panel
-                with gr.Tabs() as tabs:
-                    #---------------- tab for REPORTS SELECTION ----------------------
-
-                    with gr.Tab("Reports",elem_id = "tab-config",id = 2):
-
-                        #---------------- SELECTION METHOD - RADIO BUTTON ------------
-                        search_method = gr.Radio(
-                            choices=["Search by Report Name", "Search by Filters"],
-                            label="Choose search method",
-                            info= "Select the audit reports that you want to analyse directly or browse through categories and select reports",
-                            value="Search by Report Name",
-                        )
-
-                        #---------------- SELECT BY REPORT NAME SECTION ------------
-                        with gr.Group(visible=True) as report_name_section:
-                            # Get default report value from config if present
-                            default_report = model_config.get('app', 'dropdown_default', fallback=None)
-                            # Check if it actually exists in the master list
-                            default_report_value = [default_report] if default_report in new_report_list else None
-                            
-                            dropdown_reports = gr.Dropdown(
-                                new_report_list,
-                                label="Select one or more reports (scroll or type to search)",
-                                multiselect=True,
-                                value=default_report_value,
-                                interactive=True,
-                            )
-
-                        #---------------- SELECT BY FILTERS SECTION ------------
-                        with gr.Group(visible=False) as filters_section:
-                            #----- First level filter for selecting Report source/category ----------
-                            dropdown_sources = gr.Dropdown(
-                                ["Consolidated","Ministry, Department, Agency and Projects","Local Government","Value for Money","Thematic"],
-                                label="Select Report Category",
-                                value=None,
-                                interactive=True,
-                            )
-
-                            #------ second level filter for selecting subtype within the report category selected above
-                            dropdown_category = gr.Dropdown(
-                                [],  # Start with empty choices
-                                value=None,
-                                label = "Filter for Sub-Type",
-                                interactive=True)
-
-                            #----------- update the second level filter based on values from first level ----------------
-                            def rs_change(rs):
-                                if rs:  # Only update choices if a value is selected
-                                    return gr.update(choices=new_files[rs], value=None)  # Set value to None (no preselection)
-                                else:
-                                    return gr.update(choices=[], value=None)  # Empty choices if nothing selected
-                            dropdown_sources.change(fn=rs_change, inputs=[dropdown_sources], outputs=[dropdown_category])
-
-                         # Toggle visibility based on search method
-                        def toggle_search_method(method):
-                            """Note - this function removes the default value from report search when toggled"""
-                            if method == "Search by Report Name":
-                                # Show report selection, hide filters, and clear filter values
-                                return (
-                                    gr.update(visible=True),  # report_name_section
-                                    gr.update(visible=False),  # filters_section
-                                    gr.update(value=None),    # dropdown_sources
-                                    gr.update(value=None),    # dropdown_category
-                                    #gr.update(value=None),    # dropdown_year
-                                    gr.update()               # dropdown_reports
-                                )
-                            else:  # "Search by Filters"
-                                # Show filters, hide report selection, and clear report values
-                                return (
-                                    gr.update(visible=False), # report_name_section 
-                                    gr.update(visible=True),  # filters_section
-                                    gr.update(),              # dropdown_sources
-                                    gr.update(),              # dropdown_category
-                                    #gr.update(),              # dropdown_year
-                                    gr.update(value=[])       # dropdown_reports
-                                )
-
-                        # Pass to the event handler
-                        search_method.change(
-                            fn=toggle_search_method,
-                            inputs=[search_method],
-                            outputs=[
-                                report_name_section, 
-                                filters_section,
-                                dropdown_sources,
-                                dropdown_category, 
-                                dropdown_reports
-                            ]
-                        )
-                    
-
-                    ############### tab for Question selection ###############
-                    with gr.TabItem("Examples",elem_id = "tab-examples",id = 0):
-                        examples_hidden = gr.Textbox(visible = False)
-
-                        # getting defualt key value to display
-                        first_key = list(QUESTIONS.keys())[0]
-                        # create the question category dropdown
-                        dropdown_samples = gr.Dropdown(QUESTIONS.keys(),value = first_key,
-                                                       interactive = True,show_label = True,
-                                                       label = "Select a category of sample questions",
-                                                       elem_id = "dropdown-samples")
-                        
-                        
-                        # iterate through the questions list
-                        samples = []
-                        for i,key in enumerate(QUESTIONS.keys()):
-                            examples_visible = True if i == 0 else False
-                            with gr.Row(visible = examples_visible) as group_examples:
-                                examples_questions = gr.Examples(
-                                    QUESTIONS[key],
-                                    [examples_hidden],
-                                    examples_per_page=8,
-                                    run_on_click=False,
-                                    elem_id=f"examples{i}",
-                                    api_name=f"examples{i}",
-                                    # label = "Click on the example question or enter your own",
-                                    # cache_examples=True,
-                                )
-                            
-                            samples.append(group_examples)
-                    ##------------------- tab for Sources reporting ##------------------
-                    with gr.Tab("Sources",elem_id = "tab-citations",id = 1):
-                        sources_textbox = gr.HTML(show_label=False, elem_id="sources-textbox")
-                        docs_textbox = gr.State("")
-
-    def change_sample_questions(key):
-        # update the questions list based on key selected
-        index = list(QUESTIONS.keys()).index(key)
-        visible_bools = [False] * len(samples)
-        visible_bools[index] = True
-        return [gr.update(visible=visible_bools[i]) for i in range(len(samples))]
-
-    dropdown_samples.change(change_sample_questions,dropdown_samples,samples)
-
-
-    # ---- Guidelines Tab ----
-    with gr.Tab("Guidelines", elem_classes="max-height other-tabs"):
-        gr.Markdown("""
-        #### Welcome to Audit Q&A, your AI-powered assistant for exploring and understanding Uganda's audit reports. This tool leverages advanced language models to help you get clear and structured answers based on audit publications. To get you started, here a few tips on how to use the tool:
-        ## 💬 Crafting Effective Prompts
-        Clear, specific questions will give you the best results. Here are some examples:
-        | ❌ Less Effective | ✅ More Effective |
-        |------------------|-------------------|
-        | "What are the findings?" | "What were the main issues identified in procurement practices in the Ministry of Health in 2022?" |
-        | "Tell me about revenue collection" | "What specific challenges were identified in revenue collection at the local government level in 2021-2022?" |
-        | "Is there corruption?" | "What audit findings related to misappropriation of funds were reported in the education sector between 2020-2023?" |
-        ## ⭐ Best Practices
-        - **Be Clear and Specific**: Frame your questions clearly and focus on what you want to learn.
-        - **One Topic at a Time**: Break complex queries into simpler, focused questions.
-        - **Provide Context**: Mentioning specific ministries, years, or projects helps narrow the focus.
-        - **Follow Up**: Ask follow-up questions to explore a topic more deeply.
-        
-        ## 🔍 Utilizing Filters
-        - **Report Category & Subtype**: Use the "Reports" tab to choose your preferred report category and refine your query by selecting a specific sub-type. This will help narrow down the context for your question.
-        - **Specific Reports**: Optionally, select specific reports using the dropdown to focus on a particular document or set of documents.
-        ## 📚 Useful Resources
-                            
-        - <ins>[**Short Course: Generative AI for Everyone** (3 hours)](https://www.deeplearning.ai/courses/generative-ai-for-everyone/)</ins>
-        - <ins>[**Short Course: Advanced Prompting** (1 hour)](https://www.deeplearning.ai/courses/ai-for-everyone/)</ins>
-        - <ins>[**Short Course: Introduction to AI with IBM** (13 hours)](https://www.coursera.org/learn/introduction-to-ai)</ins>
-        Enjoy using Audit Q&A and happy prompting!
-                """)
-
-    
-
-    # static tab 'about us'
-    with gr.Tab("About",elem_classes = "max-height other-tabs"):
-        with gr.Row():
-            with gr.Column(scale=1):
-                    gr.Markdown("""The <ins>[**Office of the Auditor General (OAG)**](https://www.oag.go.ug/welcome)</ins> in Uganda, \
-                consistent with the mandate of Supreme Audit Institutions (SAIs),\
-                remains integral in ensuring transparency and fiscal responsibility.\
-                Regularly, the OAG submits comprehensive audit reports to Parliament, \
-                which serve as instrumental references for both policymakers and the public, \
-                facilitating informed decisions regarding public expenditure. 
-                
-                However, the prevalent underutilization of these audit reports, \
-                leading to numerous unimplemented recommendations, has posed significant challenges\
-                to the effectiveness and impact of the OAG's operations. The audit reports made available \
-                to the public have not been effectively used by them and other relevant stakeholders. \
-                The current format of the audit reports is considered a challenge to the \
-                stakeholders' accessibility and usability. This in one way constrains transparency \
-                and accountability in the utilization of public funds and effective service delivery. 
-                
-                In the face of this, modern advancements in Artificial Intelligence (AI),\
-                particularly Retrieval Augmented Generation (RAG) technology, \
-                emerge as a promising solution. By harnessing the capabilities of such AI tools, \
-                there is an opportunity not only to improve the accessibility and understanding \
-                of these audit reports but also to ensure that their insights are effectively \
-                translated into actionable outcomes, thereby reinforcing public transparency \
-                and service delivery in Uganda. 
-                
-                To address these issues, the OAG has initiated several projects, \
-                such as the Audit Recommendation Tracking (ART) System and the Citizens Feedback Platform (CFP). \
-                These systems are designed to increase the transparency and relevance of audit activities. \
-                However, despite these efforts, engagement and awareness of the audit findings remain low, \
-                and the complexity of the information often hinders effective public utilization. Recognizing the need for further\
-                enhancement in how audit reports are processed and understood, \
-                the **Civil Society and Budget Advocacy Group (CSBAG)** in partnership with the **GIZ**, \
-                has recognizing the need for further enhancement in how audit reports are processed and understood.   
-                
-                This prototype tool leveraging AI (Artificial Intelligence) aims at offering critical capabilities such as '
-                summarizing complex texts, extracting thematic insights, and enabling interactive, \
-                user-friendly analysis through a chatbot interface. By making the audit reports more accessible,\
-                this aims to increase readership and utilization among stakeholders, \
-                which can lead to better accountability and improve service delivery
-                
-                """)
-
-
-    # static tab for disclaimer
-    with gr.Tab("Disclaimer",elem_classes = "max-height other-tabs"):
-        with gr.Row():
-            with gr.Column(scale=1):
-                gr.Markdown("""
-                - This chatbot is intended for specific use of answering the questions based on audit reports published by OAG, Uganda for any use beyond this scope we have no liability to response provided by chatbot.
-                - The functionality and scope of this chatbot is limited to the context contained in audit reports published by OAG, Uganda in national context. 
-                - We have implemented measures to ensure the technical robustness and security of our AI system, minimizing unexpected behaviour, however we do not guarantee the full reliability, or completeness of any information provided by the chatbot and disclaim any liability or responsibility for actions taken based on its responses.
-                - The chatbot may occasionally provide inaccurate or inappropriate responses, and it is important to exercise judgment and critical thinking when interpreting its output.
-                - The use of AI within this application is transparent. When interacting with the AI, users are informed that they are engaging with an AI system
-                - The chatbot responses should not be considered professional or authoritative advice and are generated based on patterns in the data it has been trained on.
-                - The chatbot's responses do not reflect the opinions or policies of our organization or its affiliates.
-                - Any personal or sensitive information shared with the chatbot is at the user's own risk, and we cannot guarantee complete privacy or confidentiality.
-                - The chatbot is not deterministic, so there might be change in answer to same question when asked by different users or multiple times.
-                - When you use this app, we collect certain information to understand its usage, analyze performance, and continuously improve the tool for future use. This includes:
-                    - The questions you ask
-                    - The AI-generated answers
-                    - Feedback given towards to each response (in form of Thumbs-up and Thumbs-down), if any.
-                    - Usage statistics such as session duration, device type and anonymized geo-location information.
-                    
-                    We process this data based on our legitimate interest in continually enhancing the quality, security, and usability of the Audit Q&A assistant
-                - By using this chatbot, you agree to these terms and acknowledge that you are solely responsible for any reliance on or actions taken based on its responses.
-                - User can read more about the technical information about the tool in [**Readme**](https://huggingface.co/spaces/GIZ/audit_assistant/blob/main/README.md) of this tool.
-                - **This is just a prototype and being tested and worked upon, so its not perfect and may sometimes give irrelevant answers**. If you are not satisfied with the answer, please ask a more specific question or report your feedback to help us improve the system.
-                """)    
-                
-
-    def show_feedback(logs):
-        """Show feedback buttons and store logs in state"""
-        return gr.update(visible=True), gr.update(visible=False), logs 
-
-    def submit_feedback_okay(logs_data):
-        """Handle 'okay' feedback submission"""
-        return submit_feedback("okay", logs_data)
-
-    def submit_feedback_not_okay(logs_data):
-        """Handle 'not okay' feedback submission"""
-        return submit_feedback("not_okay", logs_data)
-
-    okay_btn.click(
-        submit_feedback_okay,
-        [feedback_state],
-        [feedback_row, feedback_thanks]
-    )
-    
-    not_okay_btn.click(
-        submit_feedback_not_okay,
-        [feedback_state],
-        [feedback_row, feedback_thanks]
-    )
-
-    #-------------------- Session Management + Geolocation -------------------------
-
-    # Add these state components at the top level of the Blocks
-    session_id = gr.State(None)
-    client_ip = gr.State(None)
-    
-    @demo.load(api_name="get_client_ip")
-    def get_client_ip_handler(dummy_input="", request: gr.Request = None):
-        """Handler for getting client IP in Gradio context"""
-        return get_client_ip(request)
-    
-
-    #-------------------- Gradio voodoo -------------------------
-    
-    # Update the event handlers
-    (textbox
-        .submit(get_client_ip_handler, [textbox], [client_ip], api_name="get_ip_textbox")
-        .then(start_chat, [textbox, chatbot], [textbox, tabs, chatbot], queue=False, api_name="start_chat_textbox")
-        .then(chat, 
-            [textbox, chatbot, search_method, dropdown_sources, dropdown_reports, dropdown_category, client_ip, session_id], 
-            [chatbot, sources_textbox, feedback_state, session_id], 
-            queue=True, concurrency_limit=8, api_name="chat_textbox")
-        .then(show_feedback, [feedback_state], [feedback_row, feedback_thanks, feedback_state], api_name="show_feedback_textbox")
-        .then(finish_chat, None, [textbox], api_name="finish_chat_textbox"))
-
-    (examples_hidden
-        .change(start_chat, [examples_hidden, chatbot], [textbox, tabs, chatbot], queue=False, api_name="start_chat_examples")
-        .then(get_client_ip_handler, [examples_hidden], [client_ip], api_name="get_ip_examples")
-        .then(chat, 
-            [examples_hidden, chatbot, search_method, dropdown_sources, dropdown_reports, dropdown_category,  client_ip, session_id], 
-            [chatbot, sources_textbox, feedback_state, session_id], 
-            concurrency_limit=8, api_name="chat_examples")
-        .then(show_feedback, [feedback_state], [feedback_row, feedback_thanks, feedback_state], api_name="show_feedback_examples")
-        .then(finish_chat, None, [textbox], api_name="finish_chat_examples"))
-
-    demo.queue()
-
-demo.launch()

auditchatbot_requirements.txt

@@ -1,160 +0,0 @@
-aiofiles==23.2.1
-aiohappyeyeballs==2.3.5
-aiohttp==3.10.3
-aiosignal==1.3.1
-annotated-types==0.7.0
-anyio==4.4.0
-asttokens==2.4.1
-async-timeout==4.0.3
-attrs==24.2.0
-authlib==1.3.1
-certifi==2024.7.4
-cffi==1.17.0
-charset-normalizer==3.3.2
-click==8.0.4
-comm==0.2.2
-contourpy==1.2.1
-cryptography==43.0.0
-cycler==0.12.1
-dataclasses-json==0.6.7
-datasets==2.20.0
-debugpy==1.8.5
-decorator==5.1.1
-dill==0.3.8
-exceptiongroup==1.2.2
-executing==2.0.1
-fastapi==0.112.0
-ffmpy==0.4.0
-filelock==3.15.4
-fonttools==4.53.1
-frozenlist==1.4.1
-fsspec==2024.5.0
-gradio-client==1.1.1
-gradio==4.39.0
-greenlet==3.0.3
-grpcio-tools==1.65.4
-grpcio==1.65.4
-h11==0.14.0
-h2==4.1.0
-hf-transfer==0.1.8
-hpack==4.0.0
-httpcore==1.0.5
-httpx==0.27.0
-huggingface-hub==0.23.5
-hyperframe==6.0.1
-idna==3.7
-importlib-resources==6.4.0
-ipykernel==6.29.5
-ipython==8.26.0
-itsdangerous==2.2.0
-jedi==0.19.1
-jinja2==3.1.4
-joblib==1.4.2
-jsonpatch==1.33
-jsonpointer==3.0.0
-jupyter-client==8.6.2
-jupyter-core==5.7.2
-kiwisolver==1.4.5
-langchain-community==0.0.38
-langchain-core==0.1.52
-langchain-huggingface==0.0.3
-langchain-qdrant==0.1.3
-langchain-text-splitters==0.0.2
-langchain==0.1.20
-langsmith==0.1.99
-markdown-it-py==3.0.0
-markupsafe==2.1.5
-marshmallow==3.21.3
-matplotlib-inline==0.1.7
-matplotlib==3.9.2
-mdurl==0.1.2
-mpmath==1.3.0
-multidict==6.0.5
-multiprocess==0.70.16
-mypy-extensions==1.0.0
-nest-asyncio==1.6.0
-networkx==3.3
-numpy==1.26.4
-nvidia-cublas-cu12==12.1.3.1
-nvidia-cuda-cupti-cu12==12.1.105
-nvidia-cuda-nvrtc-cu12==12.1.105
-nvidia-cuda-runtime-cu12==12.1.105
-nvidia-cudnn-cu12==9.1.0.70
-nvidia-cufft-cu12==11.0.2.54
-nvidia-curand-cu12==10.3.2.106
-nvidia-cusolver-cu12==11.4.5.107
-nvidia-cusparse-cu12==12.1.0.106
-nvidia-nccl-cu12==2.20.5
-nvidia-nvjitlink-cu12==12.6.20
-nvidia-nvtx-cu12==12.1.105
-orjson==3.10.7
-packaging==23.2
-pandas==2.2.2
-parso==0.8.4
-pexpect==4.9.0
-pillow==10.4.0
-pip==22.3.1
-platformdirs==4.2.2
-portalocker==2.10.1
-prompt-toolkit==3.0.47
-protobuf==5.27.3
-psutil==5.9.8
-ptyprocess==0.7.0
-pure-eval==0.2.3
-pyarrow-hotfix==0.6
-pyarrow==17.0.0
-pycparser==2.22
-pydantic-core==2.20.1
-pydantic==2.8.2
-pydub==0.25.1
-pygments==2.18.0
-pymupdf==1.23.26
-pymupdfb==1.23.22
-pyparsing==3.1.2
-python-dateutil==2.9.0.post0
-python-dotenv==1.0.1
-python-multipart==0.0.9
-pytz==2024.1
-pyyaml==6.0.2
-pyzmq==26.2.0
-qdrant-client==1.10.1
-regex==2024.7.24
-requests==2.32.3
-rich==13.7.1
-ruff==0.5.7
-safetensors==0.4.4
-scikit-learn==1.5.1
-scipy==1.14.0
-semantic-version==2.10.0
-sentence-transformers==3.0.1
-sentencepiece==0.2.0
-setuptools==65.5.0
-shellingham==1.5.4
-six==1.16.0
-sniffio==1.3.1
-spaces==0.29.3
-sqlalchemy==2.0.32
-stack-data==0.6.3
-starlette==0.37.2
-sympy==1.13.2
-tenacity==8.5.0
-threadpoolctl==3.5.0
-tokenizers==0.19.1
-tomlkit==0.12.0
-torch==2.4.0
-tornado==6.4.1
-tqdm==4.66.5
-traitlets==5.14.3
-transformers==4.44.0
-triton==3.0.0
-typer==0.12.3
-typing-extensions==4.12.2
-typing-inspect==0.9.0
-tzdata==2024.1
-urllib3==2.2.2
-uvicorn==0.30.6
-wcwidth==0.2.13
-websockets==11.0.3
-wheel==0.44.0
-xxhash==3.4.1
-yarl==1.9.4

model_params.cfg

@@ -6,15 +6,3 @@ TEMPERATURE = 0.2
 INFERENCE_PROVIDER = novita
 ORGANIZATION = GIZ
 
-[reader]
-TYPE = INF_PROVIDERS
-INF_PROVIDER_MODEL = meta-llama/Llama-3.1-8B-Instruct
-DEDICATED_MODEL = meta-llama/Llama-3.1-8B-Instruct
-DEDICATED_ENDPOINT = https://qu2d8m6dmsollhly.us-east-1.aws.endpoints.huggingface.cloud
-NVIDIA_MODEL = meta-llama/Llama-3.1-8B-Instruct
-NVIDIA_ENDPOINT = https://huggingface.co/api/integrations/dgx/v1
-MAX_TOKENS = 768
-INF_PROVIDER = nebius
-
-[app]
-dropdown_default = Annual Consolidated OAG 2024

utils/generator.py

@@ -74,6 +74,7 @@ def get_chat_model():
         "max_tokens": MAX_TOKENS,
     }
     
+    #### Currently the option to fetach any other Generator type are disabled #####3
     # if PROVIDER == "openai":
     #     return ChatOpenAI(
     #         model=MODEL,
@@ -110,70 +111,6 @@ def get_chat_model():
 # Initialize provider-agnostic chat model
 chat_model = get_chat_model()
 
-# ---------------------------------------------------------------------
-# Context processing - may need further refinement (i.e. to manage other data sources)
-# ---------------------------------------------------------------------
-# def extract_relevant_fields(retrieval_results: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
-#     """
-#     Extract only relevant fields from retrieval results.
-    
-#     Args:
-#         retrieval_results: List of JSON objects from retriever
-        
-#     Returns:
-#         List of processed objects with only relevant fields
-#     """
-
-#     retrieval_results = ast.literal_eval(retrieval_results)
-
-#     processed_results = []
-    
-#     for result in retrieval_results:
-#         # Extract the answer content
-#         answer = result.get('answer', '')
-        
-#         # Extract document identification from metadata
-#         metadata = result.get('answer_metadata', {})
-#         doc_info = {
-#             'answer': answer,
-#             'filename': metadata.get('filename', 'Unknown'),
-#             'page': metadata.get('page', 'Unknown'),
-#             'year': metadata.get('year', 'Unknown'),
-#             'source': metadata.get('source', 'Unknown'),
-#             'document_id': metadata.get('_id', 'Unknown')
-#         }
-        
-#         processed_results.append(doc_info)
-    
-#     return processed_results
-
-# def format_context_from_results(processed_results: List[Dict[str, Any]]) -> str:
-#     """
-#     Format processed retrieval results into a context string for the LLM.
-    
-#     Args:
-#         processed_results: List of processed objects with relevant fields
-        
-#     Returns:
-#         Formatted context string
-#     """
-#     if not processed_results:
-#         return ""
-    
-#     context_parts = []
-    
-#     for i, result in enumerate(processed_results, 1):
-#         doc_reference = f"[Document {i}: {result['filename']}"
-#         if result['page'] != 'Unknown':
-#             doc_reference += f", Page {result['page']}"
-#         if result['year'] != 'Unknown':
-#             doc_reference += f", Year {result['year']}"
-#         doc_reference += "]"
-        
-#         context_part = f"{doc_reference}\n{result['answer']}\n"
-#         context_parts.append(context_part)
-    
-#     return "\n".join(context_parts)
 
 # ---------------------------------------------------------------------
 # Core generation function for both Gradio UI and MCP

utils/retriever.py

@@ -2,8 +2,19 @@ import gradio as gr
 from gradio_client import Client
 
 
-def retrieve_paragraphs(query, country):
-    """Connect to retriever and retrieve paragraphs"""
+def retrieve_paragraphs(query:str, country:dict):
+    """Connect to retriever and retrieve paragraphs, the collection param is fixed in qdrant-client call within code.
+    
+    Params
+    ----------
+    query: the query input by user
+    country: collection 'EUDR' can filter the results based on country field. Past the values of country as str (single) or list
+    
+    Returns
+    -----------
+    Returns the top 10 retrieved>reranked results list of dict [{'answer':.....,'answer_metadata': {.....}}]
+
+    """
     try:   
 
         # Metadata selection

utils/whisp_api.py

@@ -3,7 +3,15 @@ from gradio_client import Client, handle_file
 import pandas as pd
 
 def get_value(df, colname):
-    """Fetch value from WhispAPI-style Column/Value dataframe"""
+    """Fetch value from WhispAPI-style Column/Value dataframe
+
+    Params
+    ---------
+
+    df: dataframe with statistical info from whisp api
+    colname: column name for which value need be fetched
+    
+    """
     if "Column" in df.columns and "Value" in df.columns:
         match = df.loc[df["Column"] == colname, "Value"]
         if not match.empty:
@@ -133,7 +141,8 @@ La siguiente información ha sido generada por la [WhispAPI creada por Forest Da
         return f"❌ **Analysis Error**\n\nUnable to process the geographic data: {str(e)}\n\n📋 **Troubleshooting:**\n- Verify your GeoJSON file format\n- Check file size (should be < 10MB)\n- Ensure coordinates are valid\n\nPlease try uploading again or contact support."
 
 def handle_geojson_upload(file):
-    """Handle GeoJSON file upload and call WHISP API"""
+    """Handle GeoJSON file upload and call WHISP API through the chatfed-whisp spaces. The API token is taken care by chatfed-whisp
+    space """
     if file is not None:
         try:
             # Initialize WHISP API client