DEV Community

Cover image for Building an AI-Powered SQL Chatbot With LangChain, Mistral & Streamlit
Sanjay_Balaji
Sanjay_Balaji

Posted on

Building an AI-Powered SQL Chatbot With LangChain, Mistral & Streamlit

#ai #python #langchain #nlp

(Query your PostgreSQL database using natural language)

Managing a growing database can get messy — remembering table names, joins, and Sequelize-generated column structures isn’t fun.
So I built an AI-powered SQL Chatbot that turns natural language questions into safe, optimized SQL queries, and streams results instantly.

This project uses:

  • LangChain + LangGraph
  • Mistral AI (mistral-small-latest)
  • PostgreSQL + Sequelize ORM
  • Streamlit UI

What This Chatbot Does

Simply type something like:
“Show me the last 5 orders with customer names.”
The chatbot will:
Parse the question

  • Understand the relevant tables
  • Auto-generate a safe, dialect-correct SQL query
  • Avoid soft-deleted rows
  • Perform the right JOINs using Sequelize relationships
  • Stream the result back to you in seconds
  • No manual querying. No database digging.

What This AI Chatbot Can Do

✔ Understand natural language questions
✔ Generate accurate and safe SQL
✔ Join across multiple related tables
✔ Auto-fix incorrect SQL
✔ Enforce business rules (soft-delete filters, safe SELECT-only mode)
✔ Display results interactively in Streamlit

Architecture at a Glance

  • User asks a question – “What prepaid services does customer John have?”
  • LangChain SQL Toolkit analyzes table schemas
  • Mistral LLM writes an optimized SQL query
  • PostgreSQL executes it
  • Streamlit shows the answer cleanly

Simple. Powerful. Zero SQL typing

Tech Stack

Component Purpose
Mistral AI (mistral-small-latest) Generates SQL using natural language
LangChain SQLToolkit Provides DB inspection & SQL execution
LangGraph ReAct Agent Tool-based reasoning loop
PostgreSQL Database
Streamlit Lightweight chat UI

Core Code Snippet

Here’s the essential logic behind the agent:

llm = ChatMistralAI(model="mistral-small-latest", api_key=api_key)

toolkit = SQLDatabaseToolkit(db=db, llm=llm)
tools = toolkit.get_tools()

agent = create_react_agent(llm, tools, prompt=system_prompt)

for step in agent.stream({"messages": [{"role": "user", "content": question}]}, stream_mode="values"):
    last_msg = step["messages"][-1]
    if isinstance(last_msg, AIMessage):
        st.markdown(last_msg.content)
Enter fullscreen mode Exit fullscreen mode

The SQL agent:

  • inspects schemas
  • constructs joins
  • limits output
  • blocks dangerous commands
  • retries if SQL errors Pretty smart for 20 lines of code

Why This Is Useful
This chatbot is perfect for:

  • internal dashboards
  • support teams
  • BI analysts
  • non-technical managers
  • quick debugging
  • ad-hoc data exploration

It turns your database into a chat interface, which team members can use instantly — no SQL skills required.

Try It in Your Project
If you’re building tools that query relational data, this setup is incredibly powerful.
Plug in your DB → set your rules → get a natural-language assistant instantly.

Top comments (0)