Toolkit
Reference material and runnable templates for evaluating small local language models: a cheat sheet, an Evidently observability template, and agent examples you can run in the browser.
Cheat sheet
The 23 GoF patterns plus modern service and agent patterns, and competitive-programming algorithms with C++/Python snippets, complexity budgets and step-through visualisers. Search with /.
Evidently template for local Llama agents
Datasets, versioned prompts, OpenTelemetry traces, heuristic descriptors, a local LLM judge, reports with tests, and a dashboard, all on an Ollama model. Includes a sample report.
Llama agents in the browser
Four agent patterns that GitHub Actions runs live on Ollama. You can edit and re-run them here in Python (WebAssembly) against the model's recorded replies.
Kaggle competition cheat sheets
Reference notes for the open competitions I have entered: rules, scoring, compute limits and checklists, taken from the official pages and my own runs.
ARC-AGI-2
Submission format, exact-match scoring, measured data statistics and a verified baseline scorer that shows why fixed transforms score zero.
ARC-AGI-3
Interactive games, frames and actions, the squared-efficiency score with worked examples, and the agent interface.