How lesifpe/iabadie works
IAbadie answers questions about an ADS course using a chat interface, semantic FAQ retrieval, and natural-language rewriting. The principal workflow is user question → Streamlit chat → FAISS search over FAQ embeddings → local response generation → displayed answer with the original FAQ source and links. A separate indexing workflow embeds FAQ questions into the FAISS index. The README describes Ollama, but the sampled response code instead uses llama.cpp and downloads a model from Hugging Face; the graph follows the sampled implementation. Other unsampled capabilities are not inferred.
Main entry points: search.py Semantic search · app.py Chat interface · generate_embeddings.py Embedding builder · generate_response.py Response rewriter
Components 10
Chat experience
- Chat interface
src/app.py
FAQ retrieval
- FAQ corpus
data/faq.json - Embedding builder
src/generate_embeddings.py - Sentence transformer
- FAISS index
embeddings/faq.index - Semantic search
src/search.py
Response generation
- Response rewriter
src/generate_response.py - Local LLM inference
src/generate_response.py
Other components
- Question asker
- Hugging Face model host
Connections 13
- Question asker → Chat interface: asks question
README.md - Chat interface → Semantic search: queries FAQ
src/app.py - Semantic search → FAQ corpus: reads answers
src/search.py - Semantic search → FAISS index: searches vectors
src/search.py - Semantic search → Sentence transformer: encodes query
src/search.py - Semantic search → Chat interface: returns matches
src/app.py - Chat interface → Response rewriter: rewrites answer
src/app.py - Response rewriter → Local LLM inference: generates text
src/generate_response.py - Response rewriter → Hugging Face model host: downloads model
src/generate_response.py - Chat interface → Question asker: shows answer
src/app.py - Embedding builder → FAQ corpus: reads questions
src/generate_embeddings.py - Embedding builder → Sentence transformer: encodes questions
src/generate_embeddings.py - Embedding builder → FAISS index: writes vectors
src/generate_embeddings.py