2

data analyst agent

A LangGraph-based agentic AI system that thinks like a data analyst — not just a chatbot.

GitHub Website

Description:

The MCP-Powered Data Analyst Agent is an autonomous, agentic AI system designed to perform intelligent, end-to-end data analysis on real-world datasets. Unlike traditional notebooks or dashboards, the agent reasons about the user’s question, decides what steps are necessary, and executes only the required analysis actions.

Built using a LangGraph state machine, the agent demonstrates true agentic behavior—planning, conditional execution, and grounded explanation—rather than static or prompt-only workflows.

How it works:

  1. User Input

    • User uploads a CSV or Excel dataset
    • Asks a natural-language analytical question

  2. LLM Planning

    • An LLM generates a high-level analysis plan
    • Example: inspect → analyze → visualize → explain

  3. Intent-Aware Routing

    • LangGraph determines which steps to execute
    • Simple questions skip heavy analysis
    • Insight-driven questions trigger full analysis + visualization

  4. Tool-Orchestrated Execution

    • Dataset inspection (schema, size, missing values)
    • Aggregations and computations using pandas
    • Visual insights generated with matplotlib

  5. Grounded Explanation

    • Insights are explained strictly from computed results
    • No hallucination or speculative analysis
    • Explanations adapt to student or policymaker perspectives when relevant

Why this project stands out:

  • Implements true agentic decision-making, not static pipelines
  • Avoids unnecessary computation through conditional execution
  • Demonstrates planning, reasoning, and tool orchestration
  • Works on real datasets, not toy examples
  • Focuses on correctness, explainability, and practical analytics

Tech Stack:

Python - core language
LangGraph – agent state orchestration
LangChain - tool abstraction, LLM integration, and prompt management
Google Gemini API – planning and explanation
Pandas – data analysis
Matplotlib – visualization
Streamlit – interactive UI

Project Status:

  • Core agent complete and functional
  • Supports inspection, analysis, visualization and explanation
  • Designed as a flagship agentic AI + data analytics portfolio project