As a Backend & Generative AI Engineer, I worked on a retrieval-augmented generation (RAG) platform that puts a legal team’s own corpus of judicial rulings behind natural-language search, grounding answers in the actual rulings rather than a general-purpose model’s memory.
Problem
Legal research means finding the right precedent inside a large, growing body of rulings. A general-purpose language model can answer fluently but has no guarantee its answer is grounded in an actual decision: a real risk in a domain where a wrong citation has consequences.
Solution
Retrieval-augmented generation: answers are generated from rulings retrieved from the platform’s own corpus, not from what a model happened to memorize during training.
Scale
The platform serves answers grounded in more than 10,000 judicial decisions, running in production.
My contribution
I worked on the backend: a FastAPI service exposing the retrieval and generation pipeline to the rest of the product.