JK
>LOADING // LOADING
INITIALIZING VIEW
Completedfastapipythonfaissmistral-7btogether-aisentence-transformers
LexiAI – Legal Assistant
Retrieval-Augmented Generation backend system answering legal queries with citations using FAISS, sentence-transformers, and Mistral-7B via Together API.
Features
- Natural language legal question answering
- Citation-backed responses using document chunks
- FAISS vector similarity retrieval
- PDF & DOCX preprocessing pipeline
- Structured prompt generation with metadata
Architecture
Core System Components
- Retriever module using FAISS vector search
- Generator module using Together API and Mistral-7B
- Document loader & chunker
App Layer
- FastAPI HTTP API
- Structured response formatting
- Chunk metadata tracking (filename, source)
Project Structure
app
generator.py
main.py
retriever.py
utils.py
data
index
Key Challenges
- Building accurate RAG pipeline for legal reasoning
- Maintaining citation relevance and traceability
- Managing embedding & token efficiency
Key Learnings
- RAG requires deep coordination between retrieval and prompt engineering
- Chunking strategy dramatically affects accuracy
- Metadata is crucial for real-world citation traceability