Constellation Network
September 2021 – PresentDirector of EngineeringSan Francisco
- Technical lead for a multi-year U.S. Air Force / DoD supply-chain security program: built the research prototype, then led the team delivering a platform for securing and verifying defense supply-chain data.
- Built Gate AI, a prompt-injection and jailbreak detection model for LLM applications: 97.4% F1 at a 1% false-positive rate across 16 public benchmarks (12,111 samples), with the evaluation methodology published on arXiv.
- Prototyped an LLM pipeline for HIPAA-regulated health data on Amazon Bedrock, with isolated environments for protected health information and a hybrid model combining traditional ML with LLM-derived features.
- Applied LLM compression and KV-cache optimization to cut inference memory, latency and serving cost.
- Lead the company's open-source project: roadmap, architecture, code review and releases.
