import os import threading import pandas as pd from flask import Flask, request, jsonify import gradio as gr from langchain_community.document_loaders import CSVLoader from langchain_chroma import Chroma from langchain_huggingface import HuggingFaceEmbeddings # ===================================================================== # 1. VECTOR DATABASE SETUP & INGESTION (LangChain + ChromaDB) # ===================================================================== DB_DIR = "./chroma_db" CSV_FILE = "dataset.csv" if not os.path.exists(CSV_FILE): raise FileNotFoundError(f"Missing required dataset file: {CSV_FILE}") # Initialize CPU-optimized embedding model embeddings = HuggingFaceEmbeddings( model_name="sentence-transformers/all-MiniLM-L6-v2", model_kwargs={"device": "cpu"} ) # Build or load ChromaDB vector store (handles empty directory edge cases) if not os.path.exists(DB_DIR) or not os.listdir(DB_DIR): print("Initializing ChromaDB vector store from dataset.csv...") loader = CSVLoader( file_path=CSV_FILE, metadata_columns=["intent", "action_workflow"], source_column="query" ) docs = loader.load() vectorstore = Chroma.from_documents(docs, embeddings, persist_directory=DB_DIR) print("Vector database built successfully.") else: vectorstore = Chroma(persist_directory=DB_DIR, embedding_function=embeddings) def evaluate_intent(user_query: str): """Retrieves top matching intent records from ChromaDB and computes similarity.""" if not user_query or not user_query.strip(): return "Invalid Query", "0.0%", "Please enter a non-empty query.", pd.DataFrame() # Similarity search over vector space results = vectorstore.similarity_search_with_score(user_query.strip(), k=3) if not results: return "Unknown Intent", "0.0%", "No automated workflow found", pd.DataFrame() top_doc, top_score = results[0] # Distance to similarity confidence percentage conversion confidence_raw = max(0.0, 1.0 - (top_score / 2.0)) confidence_pct = f"{round(confidence_raw * 100, 1)}%" predicted_intent = top_doc.metadata.get("intent", "Unknown") action_outcome = top_doc.metadata.get("action_workflow", "N/A") # Format grounding contexts table context_data = [ { "Matched Historical Query": doc.page_content, "Mapped Intent": doc.metadata.get("intent"), "Similarity Score": f"{round(max(0.0, 1.0 - (score / 2.0)) * 100, 1)}%", "Workflow Trigger": doc.metadata.get("action_workflow") } for doc, score in results ] return predicted_intent, confidence_pct, action_outcome, pd.DataFrame(context_data) # ===================================================================== # 2. REST API BACKEND (Flask with Robust Payload Validation) # ===================================================================== flask_app = Flask(__name__) @flask_app.route("/health", methods=["GET"]) def health(): return jsonify({"status": "active", "vector_db": "ChromaDB", "framework": "LangChain"}) @flask_app.route("/api/predict", methods=["POST"]) def api_predict(): data = request.get_json(silent=True) if not data or "query" not in data: return jsonify({"error": "Invalid request. JSON body must contain 'query' field."}), 400 query = data.get("query", "") intent, confidence, action, _ = evaluate_intent(query) return jsonify({ "query": query, "predicted_intent": intent, "confidence": confidence, "triggered_action": action }) def run_flask(): flask_app.run(host="0.0.0.0", port=5000, debug=False, use_reloader=False) # Start background API thread threading.Thread(target=run_flask, daemon=True).start() # ===================================================================== # 3. GRADIO FRONTEND DESIGN SYSTEM # ===================================================================== custom_css = """ .gradio-container { background-color: #0B0F19 !important; font-family: 'Inter', sans-serif !important; color: #F8FAFC !important; } .block, .form, div[data-testid="column"] { background: #161B26 !important; border: 1px solid #262D3D !important; border-radius: 12px !important; } h1 { color: #FFFFFF !important; font-weight: 700 !important; letter-spacing: -0.02em !important; } p, label span { color: #94A3B8 !important; } button.primary { background: linear-gradient(135deg, #635BFF 0%, #4F46E5 100%) !important; color: #FFFFFF !important; border: none !important; border-radius: 8px !important; font-weight: 600 !important; padding: 10px 20px !important; } button.primary:hover { background: linear-gradient(135deg, #4F46E5 0%, #3730A3 100%) !important; box-shadow: 0 4px 12px rgba(99, 91, 255, 0.3) !important; } textarea, input[type="text"] { background-color: #0F1420 !important; border: 1px solid #2A3347 !important; color: #F8FAFC !important; border-radius: 8px !important; } textarea:focus, input[type="text"]:focus { border-color: #635BFF !important; box-shadow: 0 0 0 2px rgba(99, 91, 255, 0.2) !important; } div[data-testid="textbox"] input { font-weight: 700 !important; color: #38BDF8 !important; } .dataframe { background-color: #0F1420 !important; color: #E2E8F0 !important; border-radius: 8px !important; border: 1px solid #262D3D !important; } .dataframe th { background-color: #1E2638 !important; color: #94A3B8 !important; font-weight: 600 !important; } .badge-status { display: inline-block; padding: 4px 12px; background: rgba(16, 185, 129, 0.1); border: 1px solid #10B981; color: #10B981; border-radius: 9999px; font-size: 12px; font-weight: 600; } """ with gr.Blocks(css=custom_css, title="E-Commerce RAG Engine") as demo: gr.HTML("""
Semantic query intent classification and automated action mapping powered by LangChain & Vector Embeddings.