I build production software and AI-powered products, from micro-SaaS tools to agentic workflows that actually ship.
- Designing and shipping AI agents with tool use, memory, and multi-step reasoning
- Building custom Skill systems for LLM applications (structured prompting, eval loops, retrieval augmentation)
- Exploring micro-SaaS opportunities in niche B2B verticals, finding distribution before writing code
- Full-stack development across web, mobile, and Chrome extensions
Capabilities:
- Agent design: tool use, memory layers, structured output, multi-turn reasoning
- Skill & prompt system development (XML-structured, eval-driven)
- RAG pipeline architecture (chunking strategies, embedding, hybrid retrieval)
- LLM API integration: Anthropic, OpenAI, local models via Ollama
- MCP server integration and custom connector development
- Evaluations and output quality frameworks for production LLMs
I build AI features the same way I build software: start with the problem, define what "working" means, then iterate toward it. For agents specifically, that means:
- Skill design first: define the task boundary, the inputs, the expected output format, and failure modes before writing a prompt
- Structured prompting: XML-tagged instructions, explicit reasoning steps, and negative examples where stakes are high
- Eval before scale: a simple correctness check beats shipping and hoping
- Human in the loop: know exactly where automation should stop and a human should confirm
