spec · lexiqe-ai · shipped · 2024
LexiQE AI
Private, local legal-document Q&A
- Role
- Sole engineer
- Domain
- RAG / NLP
- Year
- 2024
- Status
- shipped
stack →RAGFAISSFLAN-T5Sentence-TransformersStreamlit
Upload a contract or filing, ask questions in a chat, get answers grounded in the document — running entirely on-device so nothing leaves the machine.
01 · Problem
Legal documents are exactly the kind of data you cannot paste into a hosted API, but they are also dense enough that keyword search is useless.
02 · Approach
- 01PDF parsing and layout-aware chunking to keep clauses intact.
- 02FAISS index over sentence-transformer embeddings for retrieval.
- 03FLAN-T5 running locally for grounded answer generation with the retrieved passages as context.
- 04Answers cite the passage they came from so the user can verify.
03 · Outcome
- Full data privacy — no network calls after the models are downloaded.
- Context-aware answers with visible citations instead of a confident guess.