Built backend and agentic capabilities for a real-time conversational video platform.

• Shipped improvements to agent memory, greetings, and proactive-outreach behavior through production-log analysis, targeted observability tooling, and prompt evaluation
• Evaluated third-party agent-monitoring platforms for detecting and analysing undesirable agent behaviours

First full-time engineer; later led engineering across production LLM systems, infrastructure, and developer platform for enterprise e-commerce workflows

• Built evaluation systems for multi-stage LLM pipelines using customer-derived test sets and rubric-based LLM judges to measure factuality, brand voice, and policy compliance
• Developed adversarial tests and safeguards for confidential-data extraction, model-identity disclosure, prohibited-content generation, competitor mentions, malformed outputs, jailbreaks, and repeated ToS abuse
• Built prompt/output tracing and human-feedback infrastructure for failure analysis, output quality monitoring, and producing customer-specific fine-tuning datasets
• Established engineering standards and mentored 20+ engineers; cut pull-request cycles 50%, CI time 70%, and incident MTTA to under 60 seconds through automation and unified tracing, metrics, and logs

Developed Object Detection and Instance Segmentation models for use on gigapixel-scale medical images

• Technologies: Python, Keras, TensorFlow, OpenCV, OpenSlide

Developed real-time Object Detection and Tracking models for deployment on autonomous vehicles

• Technologies: Python, Keras, TensorFlow, OpenCV

• Completed ARENA exercises on linear probes and sparse autoencoders, working with model activations and learned feature representations