Tech Lead: Embeddings + LLMs

Tecton is a data platform for AI, used by Block, Coinbase, State Farm, Progressive insurance, and many others. Over 4.5 years, I led multiple foundational initiatives that shaped Tecton's infrastructure platform as we grew from Series A to Series C with an $XX million run rate. Tecton was recently acquired by Databricks.

* Led Tecton's Generative AI strategy, developing key capabilities including RAG and Agentic retrieval systems that became central features in Tecton 1.0

* Led technical design and implementation of critical infrastructure improvements including Online Store Compaction (saving customers millions in storage costs) and Advanced Aggregation features, working across batch, realtime, and streaming data infrastructure.

* On the weekly oncall pager for all 4.5 years, responding/supporting incidents 24x7x365

for tier-0 production infrastructure at 99.99% availability across 1M+ QPS

* Architected QueryTree, a query building system that became the foundation for improved query performance across multiple compute engines and enabled new aggregation patterns. Transformed a challenging technical area into one with active contributions from 10+ engineers across teams.

* Designed a Local Integration Testing infrastructure, accelerating developer iteration loop by many hours and significantly improving quality assurance across the platform

Member of the core technical leadership group at Tecton: "Tecton Architecture Group".

Distributed Systems Engineer working at the intersection of ML and systems.

Main focus was building data parallel distributed training for Determined's AI training platform (with Horovod).

I also helped maintain:

* hyper parameter optimization system built around ASHA/Hyperband

* cluster resource orchestrator + workflow execution (custom in Go)

* framework support for PyTorch, Tensorflow, and Keras

Blog post on some of my work (published after I had left): https://www.determined.ai/blog/optimizing-horovod .

Seed investment by Amplify Partners, and Series A investment by GV; Company was acquired by HP: https://venturebeat.com/business/hpe-acquires-determined-ai-to-bolster-its-high-performance-compute-business/

(kinda the face of Lyft's IPO, check the article picture!)

Team: Machine Learning Platform Team

Project: Generate real-time statistics, dashboards, and alerts on the values being read and written Lyft's internal feature service (daily peak volume of > 1 million reads/writes per minute and tier-0 [maximal availability]). Using kinesis streams for event streams, flink for computation, wavefront for metric outputs. Also connecting the kinesis streams to Druid + Superset for customer teams to be able to slice and dice their data. I owned this project fully end-to-end: project scoping, customer team interviewing, technical design (needed sign-on from two external teams), implementation, and customer team onboarding.

I also took on smaller side projects to help out the ML Platform Team: better rate limiting of feature service and support for multiple clusters for our k8s watchers for LyftLearn (Lyft's machine learning training platform).

Machine Learning at Berkeley is a student group at UC Berkeley focused on ML/AI. At the time, we had ~100 students that were members. I held various roles in the industry relations / external affairs space, including VP of External Affairs.

I started a partnership program with industry, and created the first ML/AI Career Fair at UC Berkeley. We had over 20 companies attending, and over 400 students. Notable companies participating included Lyft, NVIDIA, Nuro, and Flexport.