AI agents that trade, a protocol that clears payments, DeFi nodes that never sleep, critical minerals, and coffee from places most never visit. All running.
View my work ↓
I run several companies at once and most of the work is the same job in different clothes: find the bottleneck, build the machine that removes it, then get out of its way. A trading model, a settlement solver, a roasting profile. What drives me is the challenge — being a beginner again, learning something new, getting better every day, and doing it with people from all over the world.
I write the code myself, mostly Rust and Python, and I also know what a green coffee contract looks like in Colombia. That lets me move between founders, developers, traders and farmers without losing the plot.
AI financial intelligence for people who make their own investment decisions, built from the market data layer up.
The liability layer for payments made by AI agents.
The cardholder signs five fields once in their banking app. Every order the agent places is checked against those fields before the money moves. Inside the scope, the customer pays as if they had clicked. Outside, nothing is charged, and if it is, the agent provider is liable.
An open protocol that fits into one field of any payment rail, from Visa and Mastercard to SEPA, Wero and stablecoins. Fivefield operates the verifier.
Looking for issuers, acquirers and PSPs. If you run a rail and want to see agent payments with a liability answer, let's talk.
Join the pilot →A few things I've built, run and kept running.
A settlement solver competing in CoW Protocol's batch auctions. Finds routes, prices surplus, submits solutions, and loses money fast when the config is wrong. Written in Rust.
Staked RPC provider serving chain data to Lava's network. Endpoint operations, staking, outage recovery and chain expansion, run from my own servers.
A premium roastery with direct-trade sourcing, alongside the organic line Biokaffeerösterei Traxl. Green coffee selection, roasting, positioning and the systems behind the shop.
Co-founded trading company: a quantitative trading operation in validation, plus the import side that brings green coffee and cocoa into Europe.
Experiments and side builds. Some useful, some weird, all mine.
Five signed fields, one verifier, any rail. 107 tests, public verifier, live demo. Pilot partners wanted.
90-day validation: 10 to 15 conversations with people who run tokenized funds. Three matching problems and we build a paid pilot.
A keeper that watches your position and unwinds it before the liquidator does. Base first.
Open-source signals in, calibrated probabilities out. NORTH7 research project.
Cultivation, harvest, processing, roasting, extraction, milk, machines, workflow. Eight editions of knowledge transfer — written for producers, roasters and baristas who want to understand the whole chain.
The money isn't in issuing tokens. It's in running them afterwards. An investor redeems tokenized fund units and wants their cash: order, fund accounting, register, blockchain and payments all have to agree. When they don't, a human spends hours finding out why.
The evidence. DTCC launched its Tokenization Service on October 28 with 20 firms signed on, from JPMorgan to Citadel Securities to Fidelity. Securitize crossed $5B in tokenized assets and more than tripled its net loss year over year. Tokenization revenue down 12%. Fund administration revenue up.
The thesis. Software that finds exactly these cases, classifies them and routes them to whoever has to act. Not a dashboard with three numbers side by side. A line that says "payment confirmation missing, deadline passed, payments team is up."
What I don't know. Whether enough fund administrators have this problem, whether they'll pay, and whether the platforms just build it themselves. UBS ran a live on-chain redemption with Chainlink a year ago. The happy path is solved. My question is what happens when it doesn't apply.
The plan. No pitch. 90 days, 10 to 15 conversations with people who operate tokenized funds today. Show me your last unresolved redemption case. If three firms describe the same problem, we build a paid pilot. If not, that was a cheap lesson.
If you work in fund operations with tokenized products, I'd like to talk. Get in touch →
Why batch auctions quietly moved the price-discovery job from order books to whoever writes the best solver.
Giving AI agents money is easy. Giving them five signed fields and a liability answer is the actual product. The thinking behind Fivefield.
Both are feedback loops with a lag. Overreact to the last reading and you burn the batch.
Nodes, solvers, RPC providers and keeper bots as a one-person operation: what breaks, what pays, what to automate first.
The standard reference I kept rewriting until it stopped being wrong.
How NORTH7's four agents critique each other, and why that beats one bigger model.
I'm always looking for exciting projects with cool people around the world. A protocol that needs a solver, a market that needs a model, a product that needs someone who ships. Send it over.