How uddipanr/temporal-rag-project works
This repository implements a temporal question-answering RAG system: users submit a question, the service retrieves relevant dated corpus chunks, classifies temporal intent, adaptively reranks results for freshness, then generates an answer with source and ranking details. The corpus side includes Wikipedia ingestion, corpus preparation, chunking, and index construction; the browser demo calls the API. The sampled pipeline and API excerpts establish the query path, while some corpus-preparation connections are not evidenced in the excerpts and are shown only as unverified.
Main entry points: pipeline.py RAG Pipeline · ingest.py Article Ingestion · chunk_corpus.py Chunking · retriever.py Vector Indexing
Components 16
Corpus Preparation
- Article Ingestion
src/ingest.py - Article Corpus
src/data/corpus/articles.json - Corpus Builder
src/build_corpus.py - Chunking
src/chunk_corpus.py
Retrieval
- Vector Indexing
src/retriever.py - FAISS Index
- Semantic Retrieval
src/retriever.py
Temporal Answering
- Query Classifier
src/classifier.py - Freshness Reranker
src/reranker.py - Answer Generator
src/generator.py
API Service
- RAG Pipeline
src/pipeline.py - Query API
src/api.py
Browser Demo
- Web Demo
frontend/app.js
Other components
- Questioner
- Wikipedia API
- Groq LLM
Connections 15
- Questioner → Web Demo: submits question
frontend/app.js - Web Demo → Query API: posts query
frontend/app.js - Query API → RAG Pipeline: runs query
src/api.py - RAG Pipeline → Semantic Retrieval: retrieves documents
src/pipeline.py - RAG Pipeline → Query Classifier: classifies query
src/pipeline.py - RAG Pipeline → Freshness Reranker: reranks documents
src/pipeline.py - RAG Pipeline → Answer Generator: generates answer
src/pipeline.py - Semantic Retrieval → FAISS Index: searches index
- Vector Indexing → FAISS Index: builds index
- Chunking → Vector Indexing: supplies chunks
- Corpus Builder → Chunking: prepares text
- Article Ingestion → Wikipedia API: fetches articles
src/ingest.py - Article Ingestion → Article Corpus: writes articles
src/ingest.py - Query Classifier → Groq LLM: requests classification
src/classifier.py - Answer Generator → Groq LLM: requests completion
src/generator.py