Software Engineer building agentic evaluation pipelines and AI platform tooling — eval harnesses, MCP tooling, observability for headless coding agents, and local model-based apps for specialized fields.
Currently targeting Senior Software Engineer — Agentic / AI Platform / Evals & Tooling roles in the Bay Area.
- agent-stream — readable live terminal output for headless coding agents (Claude, Codex, Cursor, Grok): renders event streams, narrates plan/step state, and leaves a complete on-disk record (
state.json+ raw events) for every run. - Agentic evals at work — evaluation pipelines adopted across multiple engineering teams; harness design, rubric/eval design, and tooling walkthroughs. A public, sanitized showcase is in progress.
- GraphDimension — Lean 4 library for the Erdős–Harary–Tutte graph-dimension ladder, with machine-checked proofs of Erdős problem 1007 and the Frankl–Kupavskii–Swanepoel edge-threshold results (see the proof repos below).
Machine-checked proofs, no sorry: Erdos1007 · Erdos1007Dim5 · FKSEdgeThreshold · FKSProblem1 · FKSProblem2 · EdgeThresholdUniqueness
Biomedical engineering + software. Earlier research in the Jun Lab (UAB): 3 co-authored publications, first author in J Clin Med (2020).
- Resume: https://dishah3241.github.io/resume/
- LinkedIn: https://www.linkedin.com/in/dishants560
- Email: dishants560@gmail.com



