01
1 yr 3 mo
Current
NiCE
Senior AI Research Engineer
Mar 2026 – Present · Sandy, UT
- Architected a hierarchical agentic retrieval system for enterprise knowledge-base question answering, decomposing complex queries and validating retrieved context before generation to hold accuracy on multi-step procedural content — and caught a retrieval-truncation failure before it reached production.
- Owned the evaluation design, coaching logic, and live demo for a real-time, low-cost agent-coaching prototype, using statistical analysis to identify the strongest predictor of a poor call outcome and rebuilding the alerting around it without adding false alarms.
- Ran a pre-registered, falsifiable evaluation of a proposed fix for inconsistent multilingual model behaviour, disproving the leading hypothesis through controlled cross-model testing and redirecting the team to the true root cause before further engineering was spent.
- Directed the rollout of schema-validated structured outputs across a portfolio of production LLM prompts, then root-caused a platform defect where one output shape silently failed on a single model family.
- Traced a production summarization defect that would not reproduce in the standard validation environment through live logs and a controlled model-version comparison, isolating a configuration mismatch as the cause.
AI Research Engineer
Jul 2025 – Mar 2026 · Salt Lake City, UT
- Designed and architected an LLM-driven ambiguity detection system, dynamically generating clarifying questions through prompt chaining for precise, context-aware responses.
- Redesigned the prompt evaluation framework into a scalable, model-agnostic system for benchmarking across multiple LLMs using pytest and DeepEval.
- Partnered with client teams to reduce LLM hallucinations, applying prompt optimization and retrieval strategies to support a successful, low-risk go-live.