Skip to content
AI Agents

Data Analyst Agent with a Python Sandbox

Ask questions about a dataset and the agent writes and runs Python to answer.

Advanced26h1–3 people15/20 fit
TL;DR

Build Data Analyst Agent with a Python Sandbox for analysts and students in ~26h with 1–3 people. Stack: Pyodide + LLM + Next.js + Charts.

Why it wins: The agent computes real numbers, which builds trust.

Build flow

01 · Problem
Simple data questions need a coder, and plain chatbots guess numbers instead of computing them.
02 · Solution
The agent plans an analysis, writes Python that runs in a Pyodide sandbox in the browser, inspects results and iterates until it has an answer and chart.
03 · Build
  1. 1Load Pyodide with pandas
  2. 2Give the LLM a run-code tool
  3. 3Loop on errors and outputs
  4. 4Export the steps as a notebook
04 · Demo
Ask 'what predicts churn?' and show the code, chart and summary.

MVP checklist

  • Planning steps
  • Sandboxed Python
  • Auto charts
  • Reproducible notebook export

Tech stack

TechRole
PyodidePython in the browser
LLMLanguage model
Next.jsFrontend + API routes
ChartsData visualisation

Reference repos

Open-source projects you can build on. Links checked 2026-10-31.

Related ideas

More AI Agents

  • AI Agents

    ML Experiment Runner Agent

    An agent proposes, runs and compares model experiments on your dataset

    Advanced30h1–3
    PythonLLMDuckDB
    15/20

  • AI Agents

    Spreadsheet Cleanup Agent

    The agent fixes messy spreadsheets and shows every change before applying it

    Beginner10h1–2
    LLMDuckDBNext.js
    17/20

  • AI Agents

    Agent Trajectory Evaluator

    Replay agent runs, score each step and find where agents go off the rails

    Advanced26h1–3
    OpenTelemetryLLMNext.js
    15/20