Italian-first LLM-to-SQL demo application for querying a structured sales-oriented SQLite database using natural language.
The system is designed around a deterministic backend pipeline where SQL generation is fully controlled by templates and semantic models, while a local LLM is used only for conversational and semantic support.
No data is sent to external services: the application runs entirely locally using Ollama and a local model.
- Deterministic LLM2SQL pipeline (no LLM-generated SQL)
- Semantic KPI layer with synonyms and fact modeling
- Schema introspection for database awareness
- MCP-based orchestration layer
- Lightweight local execution (SQLite + local LLM)
Full documentation (Italian): ./docs/it
English documentation: planned (./docs/en)
The main documentation is currently available in Italian only.
See SETUP.md
See docs/it/getting_started.md
Additional test prompts are available in test_backend/.
This project is experimental and provided as-is.
The goal of this project is to showcase a maintainable, template-driven LLM-to-SQL architecture for BI-style use cases.
No official support, SLA or backward compatibility guarantees are provided.