SQL-Agent

An agent built to convert natural language queries to SQL queries

---

Workflow

Step-by-Step Process

1. Input:

   • User enter his query into th website
   • The query is sent to the SQL agent

2. List table from DB tool:

   • First step is to retrieve list of tables from the SQL database
   • Then the user query and the list of tables is sent to the Interpreter Agent

3. Interpreter Agent:

   • The task of the interpreter agent is to find the target table wrt the user query & the list of tables

4. Get Schema of a table Tool:

   • This tool provides the schema of the table
   • It is used to find the schema from the target table

5. Selector Agent:

   • The task of the selector agent is to find the target column wrt the user query & the schema data retrieved from the Get Schema Tool

   • If the target column is not found, then it goes to the Interpretor Agent to repeat its task

   • Else, it goes to Scriber Agent

6. Scriber Agent:

   • The task of the scriber agent is to write the SQL query baseed on the user query, target table & column table

7. Verify Agent:

   • The task of the verify agent is to check the SQL query if it is syntactically correct or not

   • Also it will check for DML statements (INSERT, UPDATE, DELETE, DROP etc.) in the SQL query

   • If any error is found in the SQL query, it will go to the Scriber Agent to regenerate the right query

   • Else it will go to the Execute Tool

8. Execute Tool:

   • This tool executes the SQL query in the database

   • If any error occured after execution, then it is sent to the Scriber Agent to regenerate the code

9. Summary Agent:

   • The task of the summary agent is to summarize the results given from the Execute Tool and the user query

---

Tech Stack

Frontend

• React : Chatbot UI

Backend

• Langchain

  • SQLDatabases Toolkit
  • Gemini LLM Integration

• Langgraph

  • To build agent workflow

• Mermaid.Ink

  • To visualize the workflow

• Mem0 (to be added)

  • To add memory to the agent

• MCP (Model Context Protocol) (to be added)

  • Will be used to serve the agent via MCP

Deployment

• Docker
  • To containerize the agent
• K8S
  • To manage containerized agents
• GCP
  • For hosting and scaling the application.

Monitoring System

• Langfuse
  • For tracking and storing agent trace logs
• Evaluation Pipeline
  • To evaluate the performance of the agent

---