AI Application Engineering
SQL Talk Agent
Business data often requires technical SQL knowledge, creating a gap between natural-language questions and usable analytical answers.
- Python
- Sentence Transformers
- FAISS
- Pandas
- SQLite
- Gemini
- RAG
- Embeddings
[ Exploration prototype — not a production system. ]
01The problem
Business data often requires technical SQL knowledge, creating a gap between natural-language questions and usable analytical answers.
02Context
I built a GenAI-powered SQL Talk Agent as an exploration into how natural-language questions can be converted into executable SQL and business-readable answers.
The prototype combines semantic retrieval, embeddings, an LLM and a SQL execution layer.
The core idea
03Ideation
The important part was not simply asking an LLM to generate SQL.
The system needed grounding so that the generated SQL was based on the available data structure and relevant context.
Before
After
04Solution
The system converts relevant information into embeddings, retrieves semantically related context and provides that context to the model before SQL generation.
- Sentence Transformers
- Embeddings
- FAISS
- Pandas
- SQLite
- Gemini
- Retrieval-Augmented Generation
05Architecture
Component 01
Question
A natural-language business question.
Hover or tap a component
06Prototype
- 01
Prototype
Built as an experimental GenAI analytics workflow.
- 02
Focus
Understanding how retrieval can improve the reliability of generated SQL rather than relying entirely on model knowledge.
07Key learning
LLMs become significantly more useful for structured-data tasks when the model is given the right context before generation.
Next exploration
AI Engineering / RAG Lab