CoreInsight
Open-source BI copilot system that unifies hybrid schema discovery, AST-level SQL guarding, a self-healing SQL loop, and an SSE-based chat experience into a single product.
Overview
I approached CoreInsight not as a classic "natural language -> SQL" demo, but as a BI copilot system closer to a production setting. The goal wasn't just to ask questions; it required bringing schema discovery, security, self-healing, and governance layers together in the same product chain.
On the backend, FastAPI, async SQLAlchemy, an SQLGlot-based guard layer, and a Qdrant-backed retrieval structure work in concert. The query flow passes through intent classification, schema discovery, SQL generation, execution, repair, and formatting stages. On the frontend, the SSE-streamed chat experience, chart generation, and admin center all meet in the same interface built with Next.js.
The resulting system focuses on the two things that are critical in LLM-based data products: accuracy and control. That's why the project isn't just a chat interface; with its auditing, semantic memory, authorization, and governance layers, it offers an open-source foundation that moves toward enterprise use.
FastAPI- and Next.js-based Text-to-SQL + BI copilot system. It turns natural language queries into safe SQL and the results into charts and analysis.
Highlights
Technical Architecture
Math Captcha Forge
MobileNetV3-based mathematical CAPTCHA solver. It unifies training, export, and the inference pipeline into a single case study.
ML Case Study