Rama Chandra User

At Handshake AI, I contribute to repository-level AI evaluation through task authoring, model evaluation, failed-task remediation, and deterministic verification. I focus on whether an agent meets a specification with reproducible evidence, not just whether its output looks successful.

On Project Dynamo, I create and refine task specifications, fixtures, expected behaviors, and pass/fail gates. I also rework tasks that did not clear review, resolving inconsistencies between prompts, repositories, tests, and verifier logic.

Through AI Evaluation Lab / RARB, my independent evaluation engineering initiative, I build frozen task contracts, acceptance gates, provenance records, and source-exact replay paths to test whether evaluation verdicts can be trusted. I compare results across Llama, Qwen, Gemma, and Mistral, and extend this work to workspace-agent evaluation through IBM Bob integration.

As an independent full-stack, AI automation, and workflow builder, I build web applications and internal tools using Next.js, React, FastAPI, and Supabase/PostgreSQL. My earlier systems and healthcare operations work included server and software setup at ChipEdge Technologies, application support, claims and billing workflows, and professional use of QuickBooks.