AI Agent Ecosystem
Multi-agent orchestration runtime with MCP-native tools, local-first inference, and durable execution.
I build production-grade systems — multi-agent runtimes, RAG pipelines, e-commerce infrastructure, and developer tooling. Local-first where it matters, type-safe end-to-end, designed for the long haul.
const engineer = {
role: "Software Engineer",
focus: ["ai-agents", "automation", "full-stack"],
stack: ["next.js", "python", "prisma", "ollama"],
status: "shipping",
}Currently
A snapshot of the systems I'm actively engineering. This list updates as focus shifts.
Multi-agent orchestration runtime with MCP-native tools, local-first inference, and durable execution.
Developer tools and CLIs that make local-first AI workflows ergonomic across Ollama, LM Studio, and OpenRouter.
Production Next.js + Prisma + PostgreSQL applications with type-safe APIs and accessible UIs.
Durable automation pipelines connecting web crawlers, AI classifiers, and external APIs.
E-commerce infrastructure with AI-assisted product discovery, search, and recommendations.
Developer tooling for prompt testing, context management, and unified model interfaces.
Capabilities
Six areas I work in regularly. Each one is backed by shipped systems, not just familiarity.
Design and ship multi-agent systems with typed contracts, durable execution, and local-first inference. RAG pipelines with hybrid retrieval and auditable citations.
Design APIs and services for correctness and scale. Event-sourced lifecycles, idempotent webhooks, row-level locking for concurrency integrity.
Type-safe, accessible, production-grade Next.js applications. Strong typing end-to-end, semantic HTML, keyboard navigation, and optimized LCP.
Durable workflow engines connecting crawlers, AI classifiers, and external APIs. Checkpointed execution, dead-letter queues, and replayable runs.
Production e-commerce backends with payment integrity, inventory locking, and AI-assisted product discovery. Designed for auditability and replay.
CLIs and IDE integrations for local-first AI development — prompt testing, context management, and unified model switching.
Selected work
Each card opens a detailed breakdown — architecture highlights, challenges solved, and links. Categories reflect the project's current state, not its importance.
Experience
Organized by domain instead of employer — the work I've done, the impact it had, and the proof points behind it.
Built agent ecosystems that orchestrate specialized agents with typed contracts, durable execution, and local-first inference. Shipped retrieval systems with hybrid search and auditable citations.
Built workflow engines that connect web crawlers, AI classifiers, and 3rd-party APIs into resilient pipelines. Designed for observability — every step is checkpointed, retried, and replayable.
Designed and shipped full-stack e-commerce platforms handling catalog, payments, and order lifecycle. Solved concurrency and overselling problems at the database layer.
Shipped client-commissioned websites and SaaS demos with strong mobile LCP, accessible navigation, and content/presentation separation. Returning-client engagement on multiple projects.
Building CLI suites and IDE integrations that make local AI development ergonomic — context management, prompt snapshot testing, and unified model switching across providers.
Skills
Grouped by domain. The list reflects what I use in production today — not everything I've ever touched.
Building agentic systems, RAG pipelines, and local-first AI tooling.
Type-safe, accessible, production React applications.
APIs and services designed for correctness and scale.
Durable pipelines that connect crawlers, AI, and external APIs.
Python-first tooling for ML, retrieval, and inference.
Relational and vector stores, modeled for query patterns.
Reproducible environments and safe deploys.
Primary languages used across recent work.
Tools and integrations that make AI development ergonomic.
Cross-platform and systems-level work.
GitHub
A live snapshot of my public GitHub presence. Pinned repos are the ones I'd point a recruiter to first.
0
39
just now
Kotlin
No description provided.
No description provided.
No description provided.
🎧 Convert .txt stories into audiobooks locally with Piper & Kokoro TTS engines — FastAPI + HTMX web UI, plugin architecture, CUDA support, one-command installers for Windows & Linux. No cloud APIs.
Android gaming without leaving Linux.
No description provided.
Writing
Long-form posts on agentic systems, RAG trade-offs, and the messy parts of shipping AI in production. The architecture is ready — posts are coming.
Upcoming pieces cover agent runtime design, hybrid retrieval trade-offs, and lessons from shipping local-first AI tools. The blog will be MDX-based with type-safe frontmatter — same architecture as the rest of this site.
content/
posts/
agent-runtime-design.mdx # draft
hybrid-retrieval-tradeoffs.mdx # draft
shipping-local-first-ai.mdx # plannedContact
If you're working on AI systems, automation, or production full-stack applications, I'd like to hear about it. Response time is usually within a day.