Senior AI Systems Engineer · AI Innovator · Researcher

Recent projects:

• Tiny AI harness + coding agent in Rust → quecto
  • a minimalist OpenAI-compatible harness and coding agent with a ~1.3 MB core and ~3.5 MB agent binary
• Parallel AI agent orchestration in cmux → antigravity-cmux-skill
  • plugin and skill for running independent Antigravity sessions in split panes, with browser automation and coordination workflows
• Local coding model benchmarking → locobench
  • practical Harbor + Ollama setup for running local agentic benchmarks and maintaining a leaderboard pipeline
• Forced-output learning skill → learn-anything-24h
  • compresses hard topics into a structured 24-hour sprint with exercises, teach-back prompts, and a final artifact

I build intelligent AI systems that combine strong engineering, modern LLM workflows, and real‑world usability.
My work focuses on context‑aware AI, LLM infrastructure, agent tooling, evaluation systems, and products that make complex workflows feel effortless.

Selected projects & research:

• Universal embedding retrieval toolkit & agentic memory layer → embenx
  • unified API for 15+ vector backends with MCP support for Claude and autonomous agents
  • pip install embenx
• Interoperable AI chat contexts → opencontext
• Academic retrieval with MCP → mcp-scholarly
• Applied model fine‑tuning in healthcare → medfit-llm
• Developer productivity automation → google-drive-upload-git-action
• Creativity‑focused Claude skills → cc-creativity-skills

I share work and thoughts on AI systems, open‑source tools, prompt design, and intelligent automation:

• Website: adityakarnam.com
• Threads: threads.com/@aditya_karnam
• Google Scholar: citations?user=WujCeDkAAAAJ

Currently exploring how to make AI systems more interpretable without sacrificing performance.

A public, read-only Model Context Protocol server exposing this portfolio's projects, recent work, research agenda, and source links as structured data for agents and assistants.

• MCP name: aditya-portfolio
• Transport: http
• URL: https://adityakarnam.com/mcp
• Discovery manifest: https://adityakarnam.com/.well-known/aditya-portfolio-mcp.json
• Health check: https://adityakarnam.com/mcp-health
• Install guide: adityakarnam.com/mcp-install

It exposes only public data: projects, posts, systems, research agenda, and source links. It does not expose private files, email, analytics, personal data, availability, compensation, immigration status, or references.

claude mcp add --transport http aditya-portfolio https://adityakarnam.com/mcp

Verify with:

claude mcp list

1. Open Claude → Settings / Customize → Connectors.
2. Choose Add custom connector.
3. Paste https://adityakarnam.com/mcp.
4. Save the connector.

Add a remote server:

• Name: aditya-portfolio
• Transport: http
• URL: https://adityakarnam.com/mcp

Clients that only support local/stdio MCP servers can bridge to this endpoint via mcp-remote:

npx mcp-remote https://adityakarnam.com/mcp

• "Use Aditya Karnam's portfolio connector. Is he a fit for this AI infrastructure role?"
• "Which projects show agent runtime or MCP experience?"
• "Give me a recruiter brief with evidence and source links."
• "Act as an engineer. Which systems should I inspect first?"