The brief
Intel’s OpenVINO team, engaged via Alten Poland, had accumulated 20+ heterogeneous LLM-evaluation report pages — each hand-built, none sharing a common template. The brief was to collapse that sprawl into a small number of reusable page templates, then build the dashboards engineers actually needed: a fast way to see how inference performance shifted from one release to the next.
A short, deliberately spec-driven engagement — strict Angular component and service specs written up front, then implemented step-by-step with GitHub Copilot and Claude Opus.
What shipped
20+ heterogeneous LLM evaluation report pages refactored into 2 reusable Angular page templates with configurable RxJS data pipelines.
~10 new composable reports enabled on top of those templates without per-report frontend rework.
Daily-diff dashboards comparing LLM inference metrics — latency, throughput, runtime parameters — across releases, giving engineers instant regression visibility versus the prior day and prior build.
A spec-driven agentic delivery workflow: written specs, then iterative implementation with GitHub Copilot and Claude Opus.