• Deployed core platform, a multi-tenant conversational and voice AI platform automating customer service for storage facility operators handling unit availability, reservations, and booking inquiries via chat and voice (LiveKit + Deepgram).

• Engineered LLM-driven intent classification with RAG-based knowledge retrieval and real-time tool execution, enabling agents to resolve customer requests end-to-end without human intervention.

• Built production observability across Langfuse, Maxim AI, and CloudWatch with automated KPI tracking cost, latency, task completion rate, and adversarial robustness ensuring SLA compliance at scale.

• Architected on AWS with PostgreSQL checkpointing and S3 Vectors, supporting per-tenant isolation, configurable business profiles, and seamless multi-business deployment from a single codebase.

Scaling web, voice and deep research agents and building other cool stuff

Meeting people

Teaching EAI6020(Infra ML/ML-Ops), ALY6000(Intro to Analytics)

• Spearheaded the development of a PySpark pipeline to enhance Fast-RCNN model performance for fish detection, resulting in a 40% increase in precision.

• Transformed 1TB of data from 500 video files into a structured CSV format, optimizing data flow and boosting training efficiency by 50% for fish surveillance initiatives.