Project 04 / RAG demo project

SourceTrace

Answers you can trace back.

Overview

A closer look
at the system.

A document Q&A demo exploring hybrid retrieval, source citations, and configurable search. Compare keyword and semantic search to see what supports each answer.

Engineering focus: Applied AI / RAG
What this demonstrates: Hybrid retrieval, source citations, configurable search, and answer evaluation.
Project context: Independent RAG demo project
Technology: LangChain / ChromaDB / BM25 / RAGAS / FastAPI / Streamlit

System study / 04Conceptual flow

01
Search documentsLangChain
02
Retrieve candidatesChromaDB · BM25
03
Weight search scoresWeighted score fusion
04
Answer with citationsRAGAS

RAG demo project

Product gallery

The interface, in context.

SourceTrace document Q&A demo with hybrid search settings, search balance, configuration comparison, and system evaluation controls — 01 / Document search demoView full size ↗

01 / The problem

Explore how retrieval choices affect document Q&A, while keeping the supporting sources available for inspection.

02 / Implementation & design

Sentence Transformer embeddings support semantic search in ChromaDB, alongside BM25 keyword matching. The retriever normalizes both scores and combines them with an adjustable weight. Users can compare hybrid, vector-only, and keyword-only retrieval; citations connect answers to the original documents.

03 / Engineering challenge

The Streamlit interface exposes retrieval settings and configuration comparisons over a FastAPI backend. RAGAS evaluation covers faithfulness, answer relevance, and context precision, with context recall when ground truth is available. The demo explores these tradeoffs without claiming a benchmark score.

Results & capabilities

Find. Verify.

Hybrid retrieval with source citations

Keyword + semantic search with adjustable weighting