Research assistant supporting Trusted CI, the NSF Cybersecurity Center of Excellence, which helps NSF-funded science projects improve their cybersecurity posture.

Advised by Prof. Barton Miller.

• Architected and deployed an async, event-driven LLM API gateway on Azure (FastAPI, Traefik, MySQL) serving 15K+ requests/day for 900+ students, with per-user rate limiting that cut API costs by 5×.
• Added OpenTelemetry tracing and Prometheus dashboards for sub-minute incident detection, cutting average resolution time by 80%.
• Hardened the Gradescope autograder with a sandboxed execution pipeline and optimized grading from O(n²) to O(n) for a 10× speedup.
• Built GitLab CI/CD pipelines with parallelized stages and one-click blue-green releases, reducing pipeline runtime by 3×.

• Built a production healthcare mobile app and voice-to-EMR pipeline that generated 500+ structured medical records at 96% field-level accuracy using ASR and agentic LLM workflows.
• Fine-tuned and deployed a LoRA-adapted Whisper ASR model (<25 WER) for clinical speech transcription.
• Built serverless RAG pipelines on AWS Lambda with Elasticsearch, surfacing 300+ clinical red flags and increasing detection of critical conditions by 470%.

• Extended the GPU query engine with string aggregation support, eliminating CPU fallback across 100% of JOB benchmark queries.
• Diagnosed and fixed 6 correctness bugs in CUDA kernels and GPU query execution, merged after review with maintainers.
• Benchmarked Sirius against DuckDB, demonstrating up to 6× faster execution on join-heavy workloads.

• Implemented Ctrl+R reverse search with an incremental search engine (input debouncing, result caching), refactoring the underlying hook into a reusable search provider.
• Designed an expandable multi-line suggestion renderer with context-aware windowing, and added snapshot tests — merged after review with Google maintainers.