📊 Analyzer GPT – Digital Data Analyzer

Turn any CSV file into insights, code, and charts using LLM-powered agents, Docker, and a Streamlit UI.

---

✨ What This Project Does

• Lets you upload a CSV file and ask natural language questions about your data.
• Uses a Data Analyzer agent to plan the analysis and generate Python code.
• Uses a Python Code Executor agent running inside Docker to safely execute the code.
• Streams back:
  • Explanations and intermediate reasoning.
  • Final insights in text form.
  • Generated visualizations as PNG images.
• Provides both:
  • A Streamlit web app for interactive exploration.
  • A CLI demo via main.py for quick testing.

---

🏗️ High-Level Architecture

LangGraph-style Overview

flowchart LR
    U[User 🧑‍💻] --> S[Streamlit UI<br/>streamlit_app.py]
    S -->|task + CSV path| T[RoundRobinGroupChat Team 🤝]

    subgraph Team
        A[Data_Analyzer_Agent 🤖] <--> C[Python_Code_Executor 🐍<br/>DockerCommandLineCodeExecutor]
    end

    T --> A
    T --> C

    C -->|PNG charts, console output| FS[(temp/ directory)]
    A -->|text insights + STOP| S
    FS -->|PNG images| S

Loading

🔧 Tech Stack

• Language: Python 3.12
• UI: Streamlit
• LLM Orchestration: autogen-agentchat, autogen-core, autogen-ext
• Code Execution: Docker via DockerCommandLineCodeExecutor
• Model Provider: OpenRouter

---

🖥️ Running the Streamlit App

From the project root:

streamlit run streamlit_app.py

Then open the URL shown in the terminal (usually http://localhost:8501).

Typical Flow in the UI

1. Upload CSV
   • Use the “Upload a CSV file” widget to select your dataset.
2. Describe Your Task
   • In the chat input, type something like:
     • “Give me summary statistics of all numeric columns.”
     • “Plot a histogram of sepal.length grouped by variety.”
3. Watch the Conversation
   • The app shows:
     • User messages 👤
     • Data Analyzer agent messages 🤖
     • Code executor outputs 👨‍💻
4. View Generated Visualizations
   • Any PNGs saved into temp/ by the code executor are displayed under Generated Visualizations.

Session state keeps:

• messages – the full chat history.
• autogen_team_state – serialized agent team state to resume conversations.
• images_shown – images that have already been displayed.

---

🧩 How the Agents Work Together

• Data_Analyzer_Agent

  • Receives your natural-language request and the context that a CSV is available.
  • Produces:
    • A short plan.
    • A single Python code block that:
      • Loads and analyzes the CSV.
      • Installs any missing packages via pip install ... if needed.
      • Saves plots to PNG files (e.g. plt.savefig('temp/some_plot.png')).
    • Post-analysis explanation, ending with STOP as a termination signal.

• Python_Code_Executor

  • Runs the code produced by the Data Analyzer inside Docker.
  • Uses WORK_DIR_DOCKER = "temp" as the working directory.
  • Produces console logs and PNG files consumed by the UI.

• RoundRobinGroupChat Team

  • Alternates between agents up to max_turns=10.
  • Uses a TextMentionTermination("STOP") condition to know when to stop.