Agentic Workflows
Multi-agent systems for technical documentation, retrieval and code review, designed around measurable task completion rather than novelty.
Applied systems work — building assistants, agents and evaluation harnesses that hold up outside the demo.
Evaluation Harnesses
Reproducible offline and online evals for LLM features, including regression suites and human-in-the-loop scoring.
Retrieval Architecture
Grounded retrieval pipelines with provenance tracking, so every generated answer can be traced back to its source.