Work from the past several years.
Led engineering for Asana's AI agent product across three teams, from private beta to general availability. We focused on reliability and core functionality, shipped GA on time, and it reached millions in add-on revenue five months later.
Owned Asana's relationships with Anthropic, OpenAI, and Mistral. The early access programs unlocked product capabilities, the team shipped one of the first production MCP servers, and we appeared in Anthropic's Claude 4 launch video.
Led the team behind Asana's RAG platform, the shared retrieval layer under Asana's AI features. It grew from one search feature to a framework handling over a million requests each workday. The platform covered vector search, context-aware document retrieval, and a cross-application integration framework.
Built Asana's LLM testing and evaluation program: an automated evaluation suite, quality benchmarks, and pre-release model testing.
Led the program that developed context engineering techniques for agentic systems. It improved quality while cutting operational cost, and the work led to a patent application.
A custom bit vector index for filtering and sorting. It cut processing time from about 7 seconds to 7 milliseconds.
Extended Asana's RAG platform to outside data sources: connectors for Google Drive, OneDrive, and Amazon Q alongside Asana-native content.