At Eskimi, I built an AI agent advisor and optimization recommendation engine with Claude Agent SDK and MCP tools, reducing AdOps campaign-investigation time from approximately 40–60 minutes to 2–3 minutes. I worked with two AdOps specialists to define rules and review scoring, diagnostics, and what-if recommendations.
I also developed an MCP server for role-based database and company-document retrieval, and expanded an OpenClaw fleet to 13 production Slack assistants. In a two-person team, I led architecture and full-stack development of an internal reporting platform.
For my IELTS scoring project, I built writing and speaking scoring APIs with BentoML and PyTorch, and trained and evaluated ModernBERT writing models and a multimodal speaking scorer. I also built the FastAPI backend and AI content pipeline for Swarmlead, and an arXiv ingestion and relevance-scoring pipeline for AI Research Radar.
