APPLIED GENAI / ENTERPRISE INTEGRATION
Natural Language to SQL
Making enterprise feature-signoff information more accessible through natural-language questions.
The challenge
Users of an enterprise feature-signoff platform needed more natural ways to interact with structured business data.
My contribution
- Built an LLM-based natural-language-to-SQL capability and backend APIs integrated with enterprise data.
- Applied retrieval-augmented generation, hybrid retrieval and reranking to support context selection.
- Incorporated in-product user feedback as a signal for improving response quality.
Architecture & approach
High-level conceptual view of the workflow:
01 · User questionNatural language
02 · AI & retrievalContext and SQL generation
03 · Enterprise dataAPI-backed response
Conceptual flow only; implementation details have been generalized.
Design considerations
- Connect model-generated interactions to existing enterprise data and API boundaries.
- Use retrieval and reranking to improve the relevance of contextual information.
- Support iterative quality improvement with user feedback.
Outcome & impact
Contributed an AI capability to the Ford Vanguard feature-signoff platform, which served approximately 1,300 users worldwide. That number describes the platform's users, not confirmed adoption of the AI feature.