I’m a Research Engineer focused on building and evaluating reliable AI systems. My work sits at the intersection of LLM evaluation, agentic systems, inference, retrieval, reinforcement learning, and production AI infrastructure.

I’m particularly interested in understanding why AI systems succeed or fail—not only at the final-answer level, but across their complete reasoning and tool-use trajectories. I build evaluation frameworks, benchmark-compression methods, reward models, golden datasets, and search-quality metrics that make model behavior measurable and improve the speed of experimentation.

Currently, I work on AI systems serving 6M+ users, where I design evaluation pipelines, multi-agent orchestration, context graphs, MCP infrastructure, hybrid retrieval systems, and RL-style experimentation workflows. My work has reduced agent failures by 28%, improved retrieval quality by 40%, and cut LLM inference costs by 70%.

My open-source research and engineering interests include:

• LLM and agent evaluation
• Long-horizon reasoning and tool-use reliability
• Benchmark compression and efficient experimentation
• Reward modeling and post-training
• RAG, retrieval quality, and context optimization
• Multimodal and voice AI systems
• AI accessibility and autonomous code remediation

I enjoy turning research ideas into reproducible systems, testing them under realistic failure conditions, and translating the results into scalable, production-ready AI infrastructure.

Research through engineering. Measure behavior, understand failure, and build systems that improve.