Wycliffe Rotich

I’m a performance-driven Software Engineer and AI Training Specialist with 2+ years of experience authoring complex codebase evaluation tasks, debugging private repositories, and executing test-driven development (TDD) pipelines across global platforms including Mercor, Turing, and Micro1. I specialise in reading large-scale, unfamiliar code architectures and constructing SWE-bench-style evaluations to stress-test frontier LLMs, with a strong focus on eliminating model technical hallucinations through RLHF workflows.

At Mercor, I’ve authored and evaluated 100+ advanced algorithmic coding tasks and backend scripts to stress-test LLM reasoning, code generation, and debugging capabilities, creating robust ground truth datasets used in SWE-bench-style model evaluations. I also constructed programmatic SFT data pairs mapping expected multi-file code modifications and runtime execution outputs to improve patch accuracy, and I benchmark model outputs against structural programming paradigms, API standards, and execution safety compliance boundaries. In parallel, I’ve delivered deep technical code reviews and fact-checking validations at Turing, designed benchmark-adopted edge-case backend prompts, and built automated test scripts that reduced manual review time by an estimated 40%—and at Micro1, I generated high-quality prompt-response pairs, ran 200+ evaluation sessions, and identified and rectified logical code hallucinations to improve RLHF loop accuracy.

Beyond evaluation and training workflows, I contribute to open-source—e.g., NeuroMesh AI Control Plane—to demonstrate hands-on ability to design and implement production-grade AI systems across distributed systems, model orchestration, and infrastructure engineering.