Skip to content
All work

P—10AI data assistant · white-labelWorking demo

Prism

Ask your business anything — and see the query behind every answer.

Type
AI data assistant
Databases
MySQL, PostgreSQL, SQL Server, SQLite
AI providers
5, including fully local
Technology
PythonFastAPISQLAlchemysqlglotChart.js
Prism answering 'Top 5 customers by total order amount' with a sentence, a bar chart and the result table

Prism is a white-label AI data assistant. A question in plain language is translated into a read-only SQL query, run against the business's own database and answered in words, with the table, a chart and the SQL shown. It also answers from uploaded documents, and it can be embedded in another product with one script tag.

01Challenge

The numbers a business needs are already in its database, but getting them out means waiting for someone who can write SQL or build a report. Pointing an AI model at production data raises the obvious worries: it might change something, show a table to the wrong person, or give a confident answer nobody can check.

02Approach

The schema is discovered automatically: tables, keys, row counts and the real values of status columns, because a guessed status in the wrong letter case silently returns nothing. Every generated query is parsed, and only a single SELECT over permitted tables runs, on a read-only session with a timeout. Each person has a table allowlist, and the model is never told that restricted tables exist.

Every answer shows its SQL, users can mark an answer right or wrong, and a benchmark script scores answers against truth queries before go-live. The model is a setting: OpenAI, Claude, Gemini, Groq, or a local model through Ollama when data may not leave the building.

Prism breaking down orders by status with a bar chart and a result table
A question in English or Bangla gets an answer, a chart and the rows behind it.

03Result

A working installation that takes its name, logo, colours and currency from configuration, with user accounts and monthly quotas, usage analytics, a full query audit log, a document knowledge base, a REST API and an embeddable chat widget — backed by 51 automated tests covering the SQL safety checks, per-user table access and sign-in.

[08]Start

Have an ideaworth building?

Let's turn it into technology people can use.