Dev Builds an "ML Team" of Agents: AutoML Pipeline From Feature Engineering to Deployment
kmeanskaran · x · 2026-10-10
Developer kmeanskaran is building a harness that turns a full ML pipeline into a team of AI agents: feed it any dataset, and agents handle data analysis, feature engineering, model training + fine-tuning, evaluation, and serving.
Key design choices: a "skeptic" agent reviews other agents' output; humans only make the important calls (choosing recommended feature processing and model selection, then promoting the best run to production); agents code in the background like engineers with visible logs; and there's a feature store, model versioning, and saved runs. Nothing is hardcoded—agents decide coding and workflow from observation and target, iterating until production. Deployment uses CI/CD + Terraform + Docker. Repo coming soon.
More from coding & agent
- Same coding agent hits 86% vs 60% SRE diagnosis accuracy once given cluster context — tianyin_xu · 2026-10-10
- CopilotKit open-sources OpenIntelligentUI: a generative UI framework for agents (2.1k stars) — aigclink · 2026-10-10
- Vibe coding has underdelivered: why waiting for the next model kills software commitment — sull · 2026-10-10
- Vibe42: zero-config mobile terminal to remotely run Claude Code on your home machine — yihui_indie · 2026-10-10
- Code-generated AI videos: 513 viral Opus 5.5 cases and prompts open-sourced, Step 5 Preview at 1/8 cost — yihui_indie · 2026-10-10
- Google DeepMind's new Polaris team pays $550k–$1.1M, no degree required — brianryhuang · 2026-10-10