Thomas Martinez

Led cross-functional pod (engineering, product, research) to evolve retrieval and synthesis pipelines for answer quality and latency. Architected multi-tenant inference paths with traffic split and automated rollbacks based on real-time quality/latency signals. Drove prompt/tooling optimization for LLM chains (routing, function calling, streaming) to reduce token usage while improving fidelity. Established evaluation & observability standards (trace IDs, metrics, error taxonomies) to triage incidents and prevent regressions. Mentored engineers via design reviews, pairing, and roadmapping; coordinated with stakeholders for quarterly delivery.

Designed a FastAPI inference gateway with request batching, timeouts, and circuit breakers for reliable, low-latency models. Productionized RAG pipelines using pgvector in PostgreSQL. Configured LLMs (fine-tuning / few-shot) for summarization, classification, and assistant workflows. Built an evaluation harness (offline/online) and mentored engineers via design reviews, pairing, and roadmap planning. Coordinated CI pipelines for automated testing and releases.

Built high-throughput .NET Core APIs for tasks, views, and permissions with strict SLAs. Optimized SQL Server queries and indexing strategies for large multi-tenant datasets. Implemented GraphQL interfaces to support scalable AI-assisted workflows; worked on frontend-backend integration and performance improvements.