Teaching a local coding agent from its own mistakes: DPO on a 30B model
The WUIC Assistant is our agentic VS Code plugin: it scaffolds Angular components, dashboards, reports, workflows, and metadata patches for apps built on the WUIC framework. Under the hood it runs qwen3-coder:30b — a Mixture-of-Experts model with ~3B active parameters — served locally through Ollama on a single RTX 4090. No API calls, no data leaving the machine. For weeks the plugin had been…
A local coding agent, the WUIC Assistant, utilizes a 30-billion-parameter model called qwen3-coder:30b through Ollama on an RTX 4090. The model runs an Angular framework, scaffolding components, reports and workflows, without any API calls. The plugin has been performing 32 tasks successfully, but a flaw was discovered in the curriculum that always passes, which no longer measures the model's ability to generalize to new prompts, nor does it indicate how often the model takes the wrong turn before correcting itself.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
