CS + Data Science @ Rutgers University New Brunswick — Go Knights!
AI agents are my jam. Building skills, orchestration systems, context engineering patterns, etc.
Right now I'm focused on how to make agents genuinely useful: what context they need, how to structure skills so they're reusable, and how to get multiple agents working together without falling apart.
AI engineering @ Chartmetric, building agent-based tools to help with music data analysis and discovery.
Personal projects:
- skill-viewer - npm package tool to spin up a clean web UI for viewing locally installed agent skills. TUI avail too!
- claude-plugins — Plugin marketplace for Claude Code — agent skills I built to streamline my own dev work. Includes spec-driven development, multi-agent orchestration, adversarial code review, and more
- agent_template — Boilerplate for a LangChain agent. Built for personal projects but I thought I'd share it here.
Agent skills and how to design them well. Lots of slop out there, how can we make them more reusable and composable?
Context engineering — what to feed an agent and when. Lately i find this ties into skill design, since skills can enable agents to be more independent in retrieving and using context.
Multi-agent orchestration, especially across different LLM providers. Check out claude-plugins for my latest experiments in this area. This skill! (Warning: it barely works for me even though i designed it for my system, but it's a fun proof of concept for now)
Claude Code. i love Claude Code, my coding agent of choice. i'm always running at least 3 instances of CC at any one time. Love the platform Anthropic has built, and i'm excited to see how it evolves.
Harness engineering. ¹ How can we make real multi-purpose agents? Claude Code is lowk a great example of this now that skills are so rampantly created.
¹ OpenAI has an interesting blog discussing harness engineering, which i consider to be the next evolution of agent design. Prompt engineering -> context engineering -> harness engineering.