7WEVSFinancial Analyst + SQL Chatbot
Ask the database a question in plain English.
A chatbot that turns plain-English questions into SQL, runs them, and shows the answer as a table with the query it used.
The live demo is paused at the moment. The walkthrough below uses sample data so you can see how it works.
- Type
- Working demo
- Role
- Design, agent, backend, interface and deployment
- Stack
- Node.js, Express, LangChain, LangGraph, Groq, OpenAI, SQL Server, Docker
Which five products sold the most last month?
- Reading the schema(order_items, products, orders)
- Writing the query
- Checking the query
- Running it
View executed query
SELECT TOP 5 p.name, SUM(oi.quantity) AS units
FROM order_items AS oi
JOIN products AS p ON p.id = oi.product_id
JOIN orders AS o ON o.id = oi.order_id
WHERE o.ordered_at >= '2026-09-01'
AND o.ordered_at < '2026-10-01'
GROUP BY p.name
ORDER BY units DESC;| Product | Units |
|---|---|
| Steel water bottle 1 L | 412 |
| Cotton tote bag | 377 |
| Bamboo toothbrush, pack of 4 | 341 |
| Glass jar 500 ml | 298 |
| Jute doormat | 254 |
The need
In most companies the numbers live in a database, and only a few people can write the SQL to get them out. Everyone else waits in a queue or works from last month's export.
What we built
A chat interface where an AI agent reads the schema, writes and checks the SQL, runs it and shows the result with the query attached.
Questions, not queries
Type a question the way you would ask a colleague. The agent reads the database schema, writes the SQL and runs it.
Checked before it runs
A LangGraph agent lists tables, inspects schemas and sample rows, validates its query before running it, and retries when the database returns an error.
You can see exactly what it did
Results come back as a table, with the executed SQL one click away, so anyone who knows SQL can verify the answer.
SQL still works
Type a query that starts with SELECT or WITH and it runs directly, skipping the AI entirely.
Company data on tap
Quotes, financial statements, net debt, dividends and splits, and news from public market data, with an AI-written summary of the headlines.
Keeps going when a provider doesn't
Model calls rotate across API keys and fall back to a second provider if the first fails.
Try it
Pick a question. The agent reads the schema, writes and checks the SQL, runs it, and shows you the query with the answer.
Ask a sample database
An illustration on a fictional shop's database. Every answer shows the SQL that produced it.
Pick a question above.
How it works
Question
Plain English
Read schema
Tables and columns
Write SQL
Agent drafts query
Check
Validate, retry on error
Run
Against the database
Answer
Table plus the query
Built so people can trust the answer
A chatbot that answers database questions is only useful if people believe it. Three decisions do most of the work:
- Show the query. Every answer carries the SQL that produced it. Someone who knows SQL can check it in seconds, and everyone else learns to trust it by seeing it hold up.
- Let the agent look before it writes. Listing tables and reading sample rows first means fewer wrong joins and invented column names.
- Fail loudly. When the database returns an error, the agent reads it and retries. If it still fails, the real error is shown, not hidden behind a vague message.
What this means for your business
The same pattern works over your sales data, inventory, admissions or patient appointments: a manager asks "how many orders did we ship late last month, by city?" and gets a table, with the query attached, without waiting for anyone.
Tech specs
- Interface
- Chat in the browser
- Agent
- LangChain SQL toolkit with a LangGraph ReAct loop
- Models
- Llama 3.3 70B on Groq, with OpenAI as fallback
- Database
- Microsoft SQL Server
- Transparency
- Executed query shown with every answer
- Market data
- Public market data: quotes, statements, dividends, news
- Backend
- Node.js and Express, with a Python agent service
- Hosting
- Docker on Hugging Face Spaces
Want something like this for your business?
Tell us what eats your team's week. You'll get a written scope with a fixed price and timeline before any work starts.