All work
/ 2024Retrieval architecture · Semantic search

LangChain RAG — Grounded QA at Scale

A retrieval-augmented QA system that grounds LLM responses in custom document corpora using Hugging Face embeddings, FAISS vector search, and OpenAI's GPT-3.5-turbo.

/ The Problem

GPT models hallucinate when asked about proprietary documents. Teams needed a way to get grounded answers over their own corpora without retraining.

/ The Approach

Chunk and embed any document set with Hugging Face embeddings, index with FAISS for fast semantic retrieval, then feed the top-k chunks to GPT-3.5-turbo with citations. A drop-in pipeline that works over any corpus without fine-tuning.

/ The Impact

Built a semantic QA pipeline with Hugging Face embeddings and FAISS retrieval feeding GPT-3.5-turbo. Answers cite source chunks. Drop-in architecture for any custom document set.

/ Highlights
  • Semantic retrieval over arbitrary document corpora with FAISS
  • Hugging Face embeddings — no proprietary embedding API required
  • Source-cited answers so every claim is traceable to a chunk
  • Drop-in architecture: point it at a new corpus and go
/ Stack
  • LangChain
  • FAISS
  • HF Embeddings
  • OpenAI
  • Python
View source on GitHub
Next project
Automated Summary Evaluator (FYP)