app.py

import os
import json
import uuid
import numpy as np
from datetime import datetime
from flask import Flask, request, jsonify, send_from_directory
from flask_cors import CORS
from werkzeug.utils import secure_filename
import google.generativeai as genai
from datasets import load_dataset
from sentence_transformers import SentenceTransformer
from transformers import pipeline
import faiss
import markdown

# Configuration
GEMINI_API_KEY = (
    "AIzaSyBbb8rH6ksakMg_v2W6hvUNzgHDI3lxWk0"  # Replace with your actual API key
)
genai.configure(api_key=GEMINI_API_KEY)

# Initialize Flask app
app = Flask(__name__, static_folder="../frontend", static_url_path="")
CORS(app)

# RAG Model Initialization
print("🚀 Initializing RAG System...")

# Load medical guidelines dataset
print("📂 Loading dataset...")
dataset = load_dataset("epfl-llm/guidelines", split="train")
TITLE_COL = "title"
CONTENT_COL = "clean_text"

# Initialize models
print("🤖 Loading AI models...")
embedder = SentenceTransformer("all-MiniLM-L6-v2")
qa_pipeline = pipeline(
    "question-answering", model="distilbert-base-cased-distilled-squad"
)

# Build FAISS index
print("🔍 Building FAISS index...")


def embed_text(batch):
    combined_texts = [
        f"{title} {content[:200]}"
        for title, content in zip(batch[TITLE_COL], batch[CONTENT_COL])
    ]
    return {"embeddings": embedder.encode(combined_texts, show_progress_bar=False)}


dataset = dataset.map(embed_text, batched=True, batch_size=32)
dataset.add_faiss_index(column="embeddings")


# Processing Functions
def format_response(text):
    """Convert Markdown text to HTML for proper frontend display."""
    return markdown.markdown(text)


def summarize_report(report):
    """Generate a clinical summary using QA and Gemini model."""
    questions = [
        "Patient's age?",
        "Patient's gender?",
        "Current symptoms?",
        "Medical history?",
    ]

    answers = []
    for q in questions:
        result = qa_pipeline(question=q, context=report)
        answers.append(result["answer"] if result["score"] > 0.1 else "Not specified")

    model = genai.GenerativeModel("gemini-1.5-flash")
    prompt = f"""Create clinical summary from:
    - Age: {answers[0]}
    - Gender: {answers[1]}
    - Symptoms: {answers[2]}
    - History: {answers[3]}
    
    Format: "[Age] [Gender] with [History], presenting with [Symptoms]"
    Add relevant medical context."""
    summary = model.generate_content(prompt).text.strip()
    print(f"Generated Summary: {summary}")  # Debugging log
    return format_response(summary)


def rag_retrieval(query, k=3):
    """Retrieve relevant guidelines using FAISS."""
    query_embedding = embedder.encode([query])
    scores, examples = dataset.get_nearest_examples("embeddings", query_embedding, k=k)
    return [
        {
            "title": title,
            "content": content[:1000],
            "source": examples.get("source", ["N/A"] * len(examples[TITLE_COL]))[i],
            "score": float(score),
        }
        for i, (title, content, score) in enumerate(
            zip(examples[TITLE_COL], examples[CONTENT_COL], scores)
        )
    ]


def generate_recommendations(report):
    """Generate treatment recommendations with RAG context."""
    guidelines = rag_retrieval(report)
    context = "Relevant Clinical Guidelines:\n" + "\n".join(
        [f"• {g['title']}: {g['content']} [Source: {g['source']}]" for g in guidelines]
    )

    model = genai.GenerativeModel("gemini-1.5-flash")
    prompt = f"""Generate treatment recommendations using these guidelines:
    {context}
    
    Patient Presentation:
    {report}
    
    Format with:
    - Bold section headers
    - Clear bullet points
    - Evidence markers [Guideline #]
    - Risk-benefit analysis
    - Include references to the sources provided where applicable
    """
    recommendations = model.generate_content(prompt).text.strip()
    references = [g["source"] for g in guidelines if g["source"] != "N/A"]
    return format_response(recommendations), references


def generate_risk_assessment(summary):
    """Generate risk assessment using the summary."""
    model = genai.GenerativeModel("gemini-1.5-flash")
    prompt = f"""Analyze clinical risk:
    {summary}
    
    Output format:
    Risk Score: 0-100
    Alert Level: 🔴 High/🟡 Medium/🟢 Low
    Key Risk Factors: bullet points
    Recommended Actions: bullet points"""
    return format_response(model.generate_content(prompt).text.strip())


# Flask Endpoints
@app.route("/upload-txt", methods=["POST"])
def handle_upload():
    """Handle text file upload and return processed data."""
    if "file" not in request.files:
        return jsonify({"error": "No file provided"}), 400

    file = request.files["file"]
    if not file or not file.filename.endswith(".txt"):
        return jsonify({"error": "Invalid file, must be a .txt file"}), 400

    try:
        content = file.read().decode("utf-8")
        if not content.strip():
            return jsonify({"error": "File is empty"}), 400

        summary = summarize_report(content)
        recommendations, references = generate_recommendations(content)
        risk_assessment = generate_risk_assessment(summary)

        response = {
            "session_id": str(uuid.uuid4()),
            "timestamp": datetime.now().isoformat(),
            "summary": summary,
            "recommendations": recommendations,
            "risk_assessment": risk_assessment,
            "references": references,
        }
        print(
            f"Response Sent to Frontend: {json.dumps(response, indent=2)}"
        )  # Debugging log
        return jsonify(response)
    except Exception as e:
        return jsonify({"error": f"Processing failed: {str(e)}"}), 500


@app.route("/")
def serve_index():
    """Serve the index.html file."""
    return send_from_directory(app.static_folder, "index.html")


@app.route("/<path:path>")
def serve_static(path):
    """Serve other static files from the frontend directory."""
    return send_from_directory(app.static_folder, path)


if __name__ == "__main__":
    app.run(host="0.0.0.0", port=5000, debug=True)