Work with me

Bring me a project. I'll tell you honestly what it'll take and what it'll cost.

What I do

Full-stack engineer and system designer. Ten years shipping production software at eBay (SE2 to MTS2), and sixteen months building agentic workflows, MCP tooling, and adversarial review systems with Claude, ChatGPT, and Claude Code. Along the way I've also shipped APIs, desktop apps, data pipelines, and a lot more across the stack. I take on projects AI or not, and I scope every engagement honestly before you commit to anything.

What I take on

I'll consider most engagement shapes, as long as the scope is something I can deliver well and the problem is interesting enough to earn my attention. I've built production AI agents, MCP servers and protocol tooling, full-stack web apps, desktop and mobile applications, APIs, data and research pipelines, graph-backed knowledge systems, and plenty more.

I've worked in domains from e-commerce and Shopify platform integrations to EEG and neuroscience data, from document processing pipelines to multi-agent safety systems. If you're not sure whether your project is a fit, ask me. I'd rather have the conversation than have you self-reject.

Three shapes of engagement

Most projects fall into one of these. If yours doesn't, reach out and we'll figure it out together.

How I scope

Every engagement is fixed scope, fixed price, agreed in writing before I start work. Here's how I get to a number:

• Size of work matters more than hours. I look at how much real thinking, writing, and building the problem takes, not at my clock.
• Risk matters. Work where I can predict the outcome is priced lower than work where I can't. If we both agree the problem is well-defined, the price reflects that.
• Leverage matters. If I can reuse patterns I've already built (CLAUDE.md specs, review pipelines, MCP tooling, session-mining plugins), you get the benefit of that leverage in the price.
• Your stakes matter. A fix on a solo builder's side project is priced differently than the same fix on a production system serving real revenue. Not because the work is different, but because the value is.
• I name my uncertainty. If I don't know how long something will take, I say so and either scope smaller or break the engagement into phases with their own numbers.
• Fixed scope, not hourly. I don't bill time and materials. It rewards slow work and punishes efficiency, and neither of us wants that.

Rough ranges

Not fixed prices. These are the ranges most engagements land in so you can tell if we're in the same ballpark before we get on a call.

• Review & Recommend: typically $1,500–$5,000 depending on system complexity and how deep you want the review
• Review & Build: typically $5,000–$20,000 depending on build scope
• Build from spec: typically $2,000–$15,000, more for larger multi-week builds. Bring me the spec and I'll give you a real number within a few days.

If your budget is below these ranges, tell me anyway. I'd rather have an honest conversation and find out we're not a fit than lose both our time to a guessing game.

About these numbers. These ranges are my current engagement prices, and they're lower than what the market typically charges for this kind of work. I'm deliberately pricing on the low end while I build out my public track record with new clients. Ten years shipping production software plus sixteen months deep in agentic development is real, but I'd rather earn the market rate than assume it. After a handful more engagements I'll raise these ranges to match the specialization. If you want to work together while the numbers are still here, now is a good moment.

What I commit to on every engagement

• Fixed scope, fixed price, in writing before I start. If the scope changes mid-engagement, we stop and rescope together. No silent expansion.
• You own everything I deliver. Code, documents, plugins, specs. Unrestricted. I retain the right to anonymize and cite patterns in public writing. You review and redact before anything goes public.
• Your data stays local unless you explicitly say otherwise. Any session history mining runs on your machine.
• Every written deliverable goes through my 4-agent review pipeline (fact-checker, technical reviewer, editorial reviewer, consistency reviewer) before you see it. Same pipeline I run on every blog post I publish.
• Honest disagreement. If I see something I think will hurt you later, I'll tell you directly and back it up with specifics. My job is to help you succeed, not to agree with you.
• No hidden uncertainty. If I don't know something, I say so. If something is outside my depth, I tell you and help you find someone better suited.
• Fair exit. Review engagements have a pro-rated exit point in the first few days. Build engagements are broken into phases, and you can stop at any phase boundary with a clean handoff. I don't lock people in.

Proof

• ShopForge — Production AI agent running a live Etsy shop end-to-end for 16 months. 34 specialized skills, three-tier memory, intent-based architecture that cut automation code 74%. The flagship proof that I build systems that survive real production, not demos. Case study →
• Reverse-engineered Cowork session format — Documented Anthropic Cowork's undocumented session storage across 146 files and mapped conversation layout across 195 sessions. This is the capability that lets me mine your Claude Code and Cowork history for patterns during a review. Session schema , conversation mapping .
• MCP API Bridge — Released, MIT licensed, 74 tests. A production-quality starter kit for turning any REST API into an MCP server. Shipped with docs, a reference implementation, and a working demo.
• zen-mcp-server — Open-source MCP server orchestrating 50+ AI models (Gemini, GPT-5, O3, Ollama) with conversation continuity across model switches. 765 commits and growing.
• Upstream bug reports — Filed and acknowledged bug reports on Google's Gemini SDK and Anthropic's Claude Code (#32561, #32213). Third-party validation that I find real problems in real tools.
• 4-agent review pipeline — The same fact-checker / technical reviewer / editorial reviewer / consistency reviewer pipeline I run on every blog post is what I'll run on your deliverables. See the process in action across 16 published posts , including self-retracted findings and corrections.
• 10 years at eBay (SE2 to MTS2) — Mobile web, View Item, Charity, Android, and the public API. Full-stack production systems serving millions of users daily. See the full timeline on the experience page .

Before you reach out

"I'm not sure which shape fits my project."

That's fine. Pick "Not sure" on the form and describe what you're working on. I'll tell you what I think.

"Is there a cost to the first meeting?"

No. If the project is a fit, we'll schedule a meeting to discuss scope. No sales funnel, no drip.

"How fast do you respond?"

Within 2 business days. If you're on a hard deadline, say so in the form and I'll prioritize.

"What if my project isn't AI-related at all?"

Still reach out. I'm a full-stack engineer first. Some of the best projects I've shipped have nothing to do with AI.

"What happens to my data?"

Anything you share during our conversations stays confidential. Anything I'm given access to during an engagement stays on your systems unless you explicitly say otherwise. I never use client work in public writing without your review and approval.

Start a conversation