Business Intelligence

Talk to Your Database
in Plain English

Ask questions on Telegram or Slack and get formatted answers, charts, and downloadable Excel reports instantly. No SQL required.

"Are you tired of waiting days for a developer just to run a simple data report?"

SupabasePostgreSQLLLMTelegram
The Problem
🧱

SQL barrier

Non-technical team members couldn't query their own data without involving a developer. Every simple question caused a delay.

Slow turnaround

Waiting for a developer to run a report meant decisions were made on stale data or gut feel rather than real numbers.

📊

No self-serve reporting

There was no way for the team to pull formatted reports or charts without custom dashboard work. Those took weeks to build.

How It Works
01

User Asks

Team member sends a plain-English question on Telegram, email, or the web interface.

02

LLM Converts

Multiple LLM calls convert the question into a safe, executable SQL query.

03

Query Runs

SQL executes against Supabase. On error, it's retried automatically with error context.

04

Answer Returned

Formatted answer with data visualizations and downloadable Excel report sent back.

Tech Stack
Supabase

Database Layer

Hosts the PostgreSQL database. Provides a secure API for querying live business data on demand with user-level access control.

  • PostgreSQL database
  • Row-level security
  • Real-time data
LLM Pipeline

NL → SQL Engine

Multiple LLM calls handle intent extraction, schema matching, SQL generation, and error recovery, producing reliable queries from free-form questions.

  • Multi-step LLM calls
  • Error retry logic
  • Schema-aware generation
Telegram / Email / Web

Multi-Platform Interface

The same pipeline runs across Telegram, email, and a web interface. Users pick whichever channel fits their workflow. The backend is shared.

  • Telegram bot
  • Email interface
  • Web chatbot
Live Example
Data BotONLINE
User
Which products had the most returns last month?
Bot
Got it. Pulling from the orders table...
Bot
✅ Top 3 by returns: 1. USB-C Hub (42) 2. Laptop Stand (31) 3. Webcam Pro (27). Full report attached as Excel.
User
Why did the USB-C Hub have so many?
Bot
Checking return reason codes… Most cited: "incompatible with MacBook" (29 of 42).

What just happened?

Natural language question received
LLM converted to SQL query
Executed against Supabase
Formatted answer returned
Excel report generated & attached
Follow-up question handled in context
Key Features
💬

Plain English Queries

Anyone on the team can ask data questions in natural language — no SQL knowledge required.

📊

Excel Report Output

Every successful query generates a downloadable, formatted Excel report alongside the text answer.

🔁

Auto Error Recovery

On a SQL error, the query is retried with error context. If still unresolved, the user is prompted to clarify.

📱

Multi-Platform

Works on Telegram, email, and the web — same intelligence, whichever channel the team prefers.

🗄️

Stored Query History

Every query and its result is stored, creating a searchable history of what was asked and when.

🔒

Secure by Design

Queries run through Supabase row-level security — users only see data they're permitted to access.

Your team deserves
self-serve data.

Works with any PostgreSQL or Supabase database. Custom to your schema, live in days.

  • Any SQL database
  • Custom to your schema
  • Live in days, not months