GenAI Engineer | Agentic AI & LLM Applications

Authored a Terminal-Bench 2 coding-agent task with an agent-facing specification, executable reference solution, deterministic Pytest verifier, and Docker environment, enabling reproducible evaluation from observable outputs.  Developed Signal Siege, a strategy-game benchmark with hidden matches and legality checks; added timeouts, process isolation, and anti-cheat safeguards to evaluate agents under controlled conditions.  Validated tasks with Harbor reference and no-op runs and GitHub Actions checks; reviewed agent trajectories and verifier edge cases to fix ambiguous requirements and false-positive results.

Evaluated and tested Amazon Nova Act agentic AI outputs against task instructions and rubrics for task completion, accuracy, reasoning quality, and instruction following; identified hallucinations, unsupported claims, and failure cases.  Reviewed AI training-data annotations and validated structured outputs against JSON Schema requirements; created reusable scoring guidelines to keep recurring model evaluations and annotation reviews consistent

Completed training across data preprocessing, model experimentation, evaluation, deployment, and monitoring; supported benchmarking and hyperparameter optimization for internal workflow automation projects.