Developer Insights: ML and analytics on src/
summary
Project Toth builds a knowledge graph of package, interpreter and OS observations, runs offline pip resolution and performance benchmarks, ingests CVEs and repo health metrics, and uses an advisor (reinforcement learning) to pick dependency versions; bots open PRs to keep stacks updated and a Thomas CLI plus a Jupyter extension (NB requirements) let developers query recommendations. They report ~110 repositories in the org and note human TensorFlow tuning can take a year for ~4–10% gains, motivating automated optimizers.
AI-written summary · about
chapters (7)
- 0:00Start
- 6:17Ask others...
- 8:07Project Thoth
- 9:06goals
- 27:00ML in Thoth - Predictor
- 39:45Example: nb-requirements
- 42:27cmd, OCP, GitHub, Notebook integration
related talks
07mInterview
24mUncovering Community and Project Insights Through Data Driven Methods
29mDevTools for the decade to come
37mMachine Learning on Source Code
39mSunset for the Wild West: Making ML disciplined by default
28mWhat is This Tech Insights Anyways?
28mDelivering AI on Code: Live Demo of source{d}
50mDeveloper Productivity – DIY (with LLM) or How I Learned to Stop Worrying and Love the AI