Academic · Python / Software Engineering
RAG document search
An academic RAG service for document questions with retrieved source passages.
- Python
- ChromaDB
- Ollama
- FastAPI
- Streamlit
- LangChain
- sentence-transformers
Project and contribution
Built an academic question-answering system over the Russian Constitution: text preparation and indexing, passage retrieval, reranking and Ollama generation. Connected the RAG pipeline to FastAPI and a Streamlit interface.
Purpose
The academic service answers questions using passages retrieved from a document. This project uses the Russian Constitution as its source. The interface displays source passages alongside the generated answer.
My work
- Prepared and chunked the text and built a ChromaDB index.
- Connected passage retrieval to cross-encoder reranking.
- Integrated answer generation through Ollama.
- Exposed the pipeline through FastAPI and connected a Streamlit interface.
Technical approach
Retrieval-augmented generation connects document search with answer generation. The system retrieves passages, reranks them and passes the context to a language model. LangChain and sentence-transformers support the processing stages.
This academic project integrates existing models and tools. Answers need to be checked against the original document.