Zacharie P.

I specialize in AI/LLM evaluation and coding-agent benchmarking, backed by 20+ years of production software engineering. Across Mercor, Snorkel AI, Aligneer, Handshake, Outlier, Turing, and Shift, my work includes evaluation harnesses, reference solutions, regression tests, grading rubrics, RLHF preference annotation, and multimodal data annotation. I build repository-based coding tasks with Docker environments and automated verifiers, then analyze agent trajectories for correctness, regressions, reward hacking, and answer leakage. In a Java coding-agent evaluation, I diagnosed API and state-integrity failures, refined task instructions, and independently verified the corrected implementation with 24 passing tests. My technical background includes Java, Python, SQL, TypeScript, React, AWS, and CI/CD, with production systems supporting 5M+ monthly requests and 99.9% uptime. I bring practical engineering judgment to model evaluation, benchmark design, and AI training data quality.