Removed the embeddings/FAISS path from the app

app.py

@@ -1,8 +1,6 @@
 from datetime import datetime
 import json
 import time
-import numpy as np
-from sentence_transformers import SentenceTransformer
 from fastapi import FastAPI, HTTPException, BackgroundTasks, Request
 from fastapi.responses import StreamingResponse, Response
 from fastapi.middleware.cors import CORSMiddleware
@@ -12,7 +10,6 @@ from llama_cpp import Llama
 from huggingface_hub import login, hf_hub_download
 import logging
 import os
-import faiss
 import asyncio
 import psutil  # Added for RAM tracking
 from google import genai
@@ -77,7 +74,6 @@ login(token=hf_token)
 
 # Models Configuration
 USE_GEMINI = os.getenv("USE_GEMINI", "false").lower() == "true"
-sentence_transformer_model = "all-MiniLM-L6-v2"
 repo_id = "unsloth/Qwen3-1.7B-GGUF" # "bartowski/deepcogito_cogito-v1-preview-llama-3B-GGUF" # "bartowski/deepcogito_cogito-v1-preview-llama-8B-GGUF"
 filename = "Qwen3-1.7B-Q4_K_M.gguf" # "deepcogito_cogito-v1-preview-llama-3B-Q4_K_M.gguf"
 
@@ -123,27 +119,6 @@ faqs = [
 ]
 
 try:
-    # Load CV embeddings and build FAISS index
-    logger.info("Loading CV embeddings from cv_embeddings.json")
-    with open("cv_embeddings.json", "r", encoding="utf-8") as f:
-        cv_data = json.load(f)
-        cv_chunks = [item["chunk"] for item in cv_data]
-        cv_embeddings = np.array([item["embedding"] for item in cv_data]).astype('float32')
-    faiss.normalize_L2(cv_embeddings)
-    faiss_index = faiss.IndexFlatIP(cv_embeddings.shape[1])
-    faiss_index.add(cv_embeddings)
-    logger.info("FAISS index built successfully")
-
-    # Load embedding model
-    logger.info("Loading SentenceTransformer model")
-    embedder = SentenceTransformer(sentence_transformer_model, device="cpu")
-    logger.info("SentenceTransformer model loaded")
-
-    # Compute FAQ embeddings
-    faq_questions = [faq["question"] for faq in faqs]
-    faq_embeddings = embedder.encode(faq_questions, convert_to_numpy=True).astype("float32")
-    faiss.normalize_L2(faq_embeddings)
-
     # Load the local model only if not using Gemini
     if not USE_GEMINI:
         logger.info(f"Loading {filename} model")
@@ -174,17 +149,6 @@ except Exception as e:
     logger.error(f"Startup error: {str(e)}", exc_info=True)
     raise
 
-def retrieve_context(query, top_k=2):
-    try:
-        query_embedding = embedder.encode(query, convert_to_numpy=True).astype("float32")
-        query_embedding = query_embedding.reshape(1, -1)
-        faiss.normalize_L2(query_embedding)
-        distances, indices = faiss_index.search(query_embedding, top_k)
-        return "\n".join([cv_chunks[i] for i in indices[0]])
-    except Exception as e:
-        logger.error(f"Error in retrieve_context: {str(e)}")
-        raise
-
 # Load the full CV at startup with explicit UTF-8 handling
 try:
     with open("cv_text.txt", "r", encoding="utf-8") as f:
@@ -656,9 +620,8 @@ async def health_check(request: Request):
 @limiter.limit("10/minute")  # Limit model info requests
 async def model_info(request: Request):
     base_info = {
-        "embedding_model": sentence_transformer_model,
-        "faiss_index_size": len(cv_chunks),
-        "faiss_index_dim": cv_embeddings.shape[1],
+        "context_source": "cv_text.txt",
+        "retrieval": "disabled",
         "tts_available": elevenlabs_client is not None,
     }
     

requirements.txt

@@ -1,14 +1,10 @@
 fastapi==0.115.0
 uvicorn==0.31.0
-sentence-transformers==3.1.1
-torch==2.4.1
-numpy==1.26.4
 huggingface_hub==0.30.1
-faiss-cpu==1.8.0
 psutil
 google-genai
 asyncio
 elevenlabs
 httpx
 llama-cpp-python==0.2.85
-slowapi==0.1.9
+slowapi==0.1.9