app.py

from fastapi import FastAPI
from pydantic import BaseModel
from typing import Dict, List
import gradio as gr
import pandas as pd
import json
from src.core import *

app = FastAPI(
    title="Insight Finder",
    description="Find relevant technologies from a problem",
)

class InputData(BaseModel):
    problem: str

class InputConstraints(BaseModel):
    constraints: Dict[str, str]
    
# This schema defines the structure for a single technology object
class Technology(BaseModel):
    """Represents a single technology entry with its details."""
    title: str
    purpose: str
    key_components: str
    advantages: str
    limitations: str
    id: int

# This schema defines the root structure of the JSON
class TechnologyData(BaseModel):
    """Represents the top-level object containing a list of technologies."""
    technologies: List[Technology]

@app.post("/process", response_model=TechnologyData)
async def process(data: InputData):
    result = process_input(data, global_tech, global_tech_embeddings)
    return {"technologies": result}


@app.post("/process-constraints", response_model=TechnologyData)
async def process_constraints(constraints: InputConstraints):
    result = process_input_from_constraints(constraints.constraints, global_tech, global_tech_embeddings)
    return {"technologies": result}


def make_json_serializable(data):
    """
    Recursively convert tensors to floats in a data structure
    so it can be passed to json.dumps.
    """
    if isinstance(data, dict):
        return {k: make_json_serializable(v) for k, v in data.items()}
    elif isinstance(data, list):
        return [make_json_serializable(item) for item in data]
    elif isinstance(data, tuple):
        return tuple(make_json_serializable(item) for item in data)
    elif hasattr(data, 'item'):  # torch.Tensor with single value
        return float(data.item())
    else:
        return data
        
def process_input_gradio(problem_description: str):
    """
    Processes the input problem description step-by-step for Gradio.
    Returns all intermediate results.
    """
    # Step 1: Set Prompt
    prompt = set_prompt(problem_description)

    # Step 2: Retrieve Constraints
    constraints = retrieve_constraints(prompt)

    # Step 3: Stem Constraints
    constraints_stemmed = stem(constraints, "constraints")
    save_dataframe(pd.DataFrame({"stemmed_constraints": constraints_stemmed}), "constraints_stemmed.xlsx")

    # Step 4: Global Tech (already loaded, just acknowledge)
    # save_dataframe(global_tech_df, "global_tech.xlsx") # This is already done implicitly by loading

    # Step 5: Get Contrastive Similarities
    result_similarities, matrix = get_contrastive_similarities(
        constraints_stemmed, global_tech, global_tech_embeddings
    )
    save_to_pickle(result_similarities)

    # Step 6: Find Best List Combinations
    best_combinations = find_best_list_combinations(constraints_stemmed, global_tech, matrix)

    # Step 7: Select Technologies
    best_technologies_id = select_technologies(best_combinations)

    # Step 8: Get Technologies by ID
    best_technologies = get_technologies_by_id(best_technologies_id, global_tech)

    # Format outputs for Gradio
    matrix_display = matrix #.tolist() # Convert numpy array to list of lists for better Gradio display

    print(best_combinations)
    
    result_similarities_display = {
        item['id2']: f"{item['constraint']['title']} ({item['similarity'].item():.3f})"
        for item in result_similarities
    }
    
    
    # Convert to JSON-safe format
    safe_best_combinations = make_json_serializable(best_combinations)
    safe_best_technologies = make_json_serializable(best_technologies)
    
    # Now this will work safely:
    best_combinations_display = json.dumps(safe_best_combinations, indent=2)
    best_technologies_display = json.dumps(safe_best_technologies, indent=2)

    print(f"constraints_stemmed : {constraints_stemmed}")
    
    return (
        prompt,
        ", ".join(constraints),
        ", ".join(constraints_stemmed),
        "Global technologies loaded and ready.", # Acknowledge tech loading
        str(result_similarities_display), # Convert dict to string for display
        pd.DataFrame(matrix_display, index=constraints_stemmed, columns=global_tech['name']), # Display matrix as DataFrame
        best_combinations_display,
        ", ".join(map(str, best_technologies_id)),
        best_technologies_display
    )

# --- Gradio Interface Setup ---

# Define the input and output components
input_problem = gr.Textbox(
    label="Enter Problem Description",
    placeholder="e.g., Develop a secure and scalable e-commerce platform with real-time analytics."
)

output_prompt = gr.Textbox(label="1. Generated Prompt", interactive=False)
output_constraints = gr.Textbox(label="2. Retrieved Constraints", interactive=False)
output_stemmed_constraints = gr.Textbox(label="3. Stemmed Constraints", interactive=False)
output_tech_loaded = gr.Textbox(label="4. Global Technologies Status", interactive=False)
output_similarities = gr.Textbox(label="5. Result Similarities (Constraint -> Top Technologies)", interactive=False)
output_matrix = gr.Dataframe(label="6. Similarity Matrix (Constraints vs. Technologies)", interactive=False)
output_best_combinations = gr.JSON(label="7. Best Technology Combinations Found")
output_selected_ids = gr.Textbox(label="8. Selected Technology IDs", interactive=False)
output_final_technologies = gr.JSON(label="9. Final Best Technologies")

# Create the Gradio Blocks demo
with gr.Blocks() as gradio_app_blocks:
    gr.Markdown("# Insight Finder: Step-by-Step Technology Selection")
    gr.Markdown("Enter a problem description to see how relevant technologies are identified through various processing steps.")
    input_problem.render()
    process_button = gr.Button("Process Problem")
    
    with gr.Column():
        output_prompt.render()
        output_constraints.render()
        output_stemmed_constraints.render()
        output_tech_loaded.render()
        output_similarities.render()
        output_matrix.render()
        output_best_combinations.render()
        output_selected_ids.render()
        output_final_technologies.render()

    process_button.click(
        fn=process_input_gradio,
        inputs=input_problem,
        outputs=[
            output_prompt,
            output_constraints,
            output_stemmed_constraints,
            output_tech_loaded,
            output_similarities,
            output_matrix,
            output_best_combinations,
            output_selected_ids,
            output_final_technologies
        ]
    )

gr.mount_gradio_app(app, gradio_app_blocks, path="/gradio")