Ask questions about a dataset and the agent writes and runs Python to answer
Advanced26h1–3
PyodideLLMNext.js
15/20ML Experiment Runner Agent
An agent proposes, runs and compares model experiments on your dataset.
Advanced30h1–3 people15/20 fit
Build ML Experiment Runner Agent for kaggle competitors and ml students in ~30h with 1–3 people. Stack: Python + LLM + DuckDB + Charts.
Why it wins: Turns AutoML into a transparent, explainable agent.
Build flow
01 · Problem
Trying many models and features by hand is slow and poorly tracked.
02 · Solution
The agent profiles the data, proposes experiments, runs them in a sandbox, logs metrics and explains what worked.
03 · Build
- 1Profile the dataset
- 2Generate experiment configs
- 3Run training jobs
- 4Summarise results
04 · Demo
Run five experiments on a public dataset and show the leaderboard the agent built.
MVP checklist
- Experiment proposals
- Sandboxed runs
- Metric leaderboard
- Explanations
Tech stack
| Tech | Role | |
|---|---|---|
| Python | Backend / scripting | Backend |
| LLM | Language model | AI |
| DuckDB | In-process analytics SQL | Data |
| Charts | Data visualisation | UI |
Related ideas
More AI AgentsAI Agents
Multi-Agent Research Brief
Produces a one-page cited brief on any question in minutes
Advanced36h2–4PythonLangGraphSearch API16/20AI Agents
Web Form Filling Agent
Completes repetitive multi-step web forms from a short instruction
Advanced48h2–4PythonPlaywrightLLM15/20