Senior Software Engineer at Microsoft · Azure AI Search · Redmond, WA

I build retrieval systems that help enterprise AI find the right evidence at billion-vector scale. I'm the technical lead for a team of five engineers and scientists building a new vector-search diversity capability, owning technical direction and delivery from algorithm analysis and distributed architecture through cross-team alignment and hands-on implementation.

Website & interactive demos · Experience · LinkedIn

• Agentic retrieval: Led filter and boost generation from research to production, turning open-ended intent into a bounded, verifiable operator set. Aligned research and product teams and shipped improvements to Public Preview.
• Benchmarking & production delivery: Architected agentic-retrieval benchmarking across CPU, memory, throughput, and latency; proposed and shipped its production billing model for a new serverless enterprise search offering.
• Vector search & performance: Drove vector quantization from Public Preview to general availability, helping deliver 8–32× customer cost savings and up to 20× lower latency. Optimized distance kernels with SIMD, loop unrolling, independent accumulators, and fused multiply-add.
• Search relevance & reliability: Designed hybrid-search subscore fusion, resource-aware HNSW quota enforcement that reduced limit overshoot by 100×, and an extensible facet-aggregation engine built on a custom lexer, parser, and evaluator.

I care about performance, correctness, and distributed systems — from reviewing vector algorithms and diagnosing cross-team production incidents to mentoring engineers and grounding LLM agents in governed enterprise knowledge.

Languages: C++, C#, Java, Python, TypeScript.

• My website: a scripted chat experience and interactive illustrations of vector diversity, agentic retrieval, quantization, SIMD, and HNSW.
• Bad Designs Museum: a deliberately questionable collection of terrible web UX.
• Harness engineering & agent memory: exploring how agents organize context and carry useful knowledge across tasks.

• Human pose estimation: Led a student team training a heatmap-based 2D pose model from scratch on COCO 2017. Architected the model, cloud training pipeline, data augmentation, visualization, and deployment.
• Semantic road segmentation: Built a U-Net training pipeline on KITTI Road with Keras, custom losses, and augmentation, achieving up to 99.1% F1.
• Self-supervised monocular depth estimation: Adapted Monodepth2 to TensorFlow and 136,000 DrivingStereo image pairs, learning depth without ground-truth depth labels.
• Embedded C optimization: Implemented and optimized a discrete cosine transform, exploring algorithmic, software, and hardware-level performance improvements.
• Battlesnake reinforcement learning: Trained an agent through self-play and games against public opponents.

I mentor University of Victoria software engineering students on requirements, scope, and evaluation for production AI challenges. Previously, I:

• Founded a seniors' digital literacy program that reached 650+ seniors with 180 volunteers across 30 workshops, and continued under successors.
• Co-led the UVic IEEE Student Branch, building a 14-workshop technical skills series reaching 350+ students.
• Founded and chaired a 200+ attendee technology and business strategy conference.

I graduated from the University of Victoria in Electrical & Computer Engineering, specializing in computational intelligence, with a 97% cumulative average and 20+ scholarships and awards worth over $100K. I'm a Schulich Leader Scholar and was selected for Y Combinator's inaugural AI Startup School in 2025.

Reach out on LinkedIn to connect or collaborate.