First public repos

status

closed · win

when

2026-02 → 2026-02

came from

Team enablement

led to

Agent Zone, Dream Team, SnapDragon-Coder, Bench lane, Greenfield DevOps platform

February 2026: local models via Ollama, a deterministic code-generation benchmark across 22 models, and Temporal patterns in Go.

Why it started

Before the organisation had its own licensing, evenings and weekends went to the questions the day job raised but could not answer on the clock: how good are local models, really, at the code I write? And what does a workflow engine buy me that a shell script does not?

What it is

Four repos, all public on GitHub:

• local-llm and local-llm-examples: multi-file code analysis tools driven by local models through Ollama, in Python and Go.
• specgen: a deterministic benchmark of code-generation accuracy across 22 models and 20 levels of Go coding task. The first time I measured a model instead of trusting it.
• temporal-examples: Temporal workflows in Go covering dependency injection, idempotency, container lifecycle, signals and cross-cluster bridges.¹

Where it stands

Closed as repos; alive as habits. specgen’s “frozen task set, every model, every level” is the ancestor of the frozen-77 benchmark in SnapDragon-Coder and of the bench lane. temporal-examples became the rule that every multi-hour job in the lab is a workflow (Temporal or it’s a POC).

Lineage

• Came from: team enablement
• Led to: Agent Zone, SnapDragon-Coder, bench lane, greenfield platform