At PureCode AI, I lead technical direction for VS Code Assistant and Cloud Agent with an 18-person cross-functional team. I translate developer needs into architecture, feasibility, scope, and UX.
I raised internal task success from 82% to 90–95% through context engineering, and evaluate model and prompt changes in Langfuse with reasoning-model LLM judges. I also improved token-cost efficiency by an internally estimated 3x through prompt caching, stale-context pruning, and automatic summarization.
At PureCode AI, I architected and prototyped specialized subagents, shared progress coordination, and file-based memory. I also shaped hybrid retrieval and designed tool permissions and file protections for developer control.
Previously, at Ovonts, I architected campaign analytics and monitoring for ROI analysis, plus an influencer discovery module for campaign targeting. At 1st Main, I built React interfaces and integrated APIs and services; I also built projects including Subagent Bridge, Bookmark AI, God MCP, and Project Orbit AI.
