Show HN: QueryPanel – AI Driven Dashboards
querypanel.ioI’m Csaba, and I built QueryPanel (querypanel.io) — a server-side SDK that turns natural language into SQL and helps products offer AI-driven dashboards to their users.
Many tools want to let users “chat with their data,” but building this internally means dealing with schema extraction, LLMs, embeddings, prompt engineering, and accuracy tuning. QueryPanel handles these parts for you.
What it does? - Discovers your DB schema automatically - Generates SQL from natural language - Uses embeddings + LLM to improve accuracy - Includes an admin UI to add golden queries and annotate columns - Generates a chart definition that you can use for generating charts with your favourite chart builder
Where it’s useful?
If your product has dashboards or analytics, QueryPanel lets your users customize them through text (“create a chart of revenue by country last quarter”).
Security
QueryPanel does not store credentials or user data — only the schema. All execution happens on your side.
Why I built it?
I kept seeing the same problem repeated: everyone wants NL → SQL features. So I abstracted it.
Demo, feedback, and criticism all welcome. I'm especially curious about edge cases and how people think about reliability in NL → SQL systems. Happy to answer questions here. Some extra details: 1. How accuracy is handled
QueryPanel uses a hybrid approach: schema embeddings for semantic column/table matching LLM reasoning for query construction optional “golden queries” to anchor common patterns
If the model is unsure, it returns multiple SQL candidates with confidence scores. 2. Supported databases
PostgreSQL and ClickHouse are stable. 3. Deployment model
Everything is server-side.
You keep your data + credentials.
QueryPanel only keeps: table names, columns, and metadata you enrich in the admin UI. 4. Why not build a full dashboard product?
Some comments usually ask this.
The idea is that most products already have UI — they just don’t want to build the LLM-to-SQL layer. Keeping it as an SDK makes it flexible for SaaS tools, internal dashboards, BI, etc. 5. What I’m looking for People who want to test NL → SQL for their existing analytics Feedbacks Suggestions for benchmark datasets to validate accuracy Thanks to anyone who takes a look, honest feedback is very welcome.