Project · GM MEXICO
LucIA
Hybrid-search RAG co-pilot for GM's commercial teams. −40% hallucinations, −60% time-to-insight.
LucIA is a RAG co-pilot I shipped for GM Mexico’s commercial teams — a system that lets people ask questions in plain language over the documents and data they work with every day, and get grounded answers instead of a search-results page.
Outcomes
- −40% hallucinations versus the baseline retrieval approach, driven by a hybrid-search retrieval architecture.
- −60% time-to-insight for the commercial teams using it.
The retrieval layer combines semantic and keyword search (hybrid search) so answers stay grounded in the underlying documents rather than the model’s imagination.
Details here are intentionally outcome-focused — the architecture specifics belong to GM. If you want to talk about how systems like this are built, the deep technical story is on the Martita page, which is fully mine.