At AnswersNow, I built an AI session documentation service that transcribes therapy sessions with Whisper and drafts structured notes through AWS Bedrock for BCBA review. I also designed a retrieval-augmented generation pipeline for personalized learning modules using clinical program templates and each child's interests.
I built an LLM gateway and evaluation harness to manage model routing, prompt changes, and regression testing before release. I also implemented PHI safeguards and human-in-the-loop workflows that capture clinician edits as feedback.
At Modal, I built components of the GPU worker scheduler and worked on reducing model inference cold starts. I also developed Python SDK features and reference deployments for LLM serving, transcription, and embedding generation.
Earlier, I built SaaS API and database replication connectors at Fivetran, then developed NLP, forecasting, and computer vision models at NLP Logix. My work has spanned the ML lifecycle, from data pipelines and modeling through deployment, monitoring, and evaluation.
