At Surge AI, I create C++ and Python programming problems for LLM coding benchmarks, spanning graph algorithms, dynamic programming, data structures, and complexity analysis.
I review model-generated code, trace reasoning failures, and design adversarial inputs that expose hidden errors. I also build checkers, validators, and input generators that recognize valid approaches while catching incomplete or inefficient solutions.
At Encord, I work on Python and FastAPI services for Encord Annotate, supporting dataset imports, AI-assisted labeling, reviewer assignments, and versioned training-data exports. For Encord Active and Encord Index, I've built containerized ML and data pipelines for evaluation, retraining, web extraction, normalization, deduplication, and validation.
I use PostgreSQL, Kafka, Redis, and Docker to coordinate distributed workloads, with testing, monitoring, and recovery built into data-intensive systems.
