Players put real money behind a match. Escrow holds every stake until a result is settled, wallet deposits and withdrawals run across multiple gateways and crypto, and every transaction is logged so a retry can never double pay.
Players stake real money on a match they play themselves. Funds cannot move on anyone's say-so, and a dropped connection cannot become a double payout.
What we built
Escrow holds every stake until a result is settled. Tournament lifecycle and capacity checks, winner voting with admin override, multi currency wallets with card and crypto deposits, idempotent transaction logging, threshold based KYC, and real time chat with moderation.
What it taught us
Every payment path was called more than once in the wild. The retry safety designed in up front is the reason there was never a duplicate payout to unwind.
A private AI software firm that runs on one laptop. Seventeen departments over one execution queue and one audit trail, local open source models doing the everyday work, and no cloud spend without a named human approving it.
internal system · built and used in house, not sold
React 19 · FastAPI · Postgres · Ollama · Docker
The constraint
Most AI products are one assistant answering one question at a time, hosted on somebody else's hardware, with your data going along for the ride.
What we built
Seventeen departments, each a lens over one execution queue, one audit trail, one chat system and one approval inbox. Slash commands dispatch real work: proposing diffs, running toolchains in containers, driving a site in a sealed browser, building real office documents. Nothing ships unseen and nothing spends without a name.
What it taught us
The model is not the moat. Memory, project ledgers, sandboxes, review discipline and cost governance compound the longer the system runs, and the model behind them is a replaceable part.
Buyer and seller funds sit in escrow against milestones, payments clear through several gateways, KYC steps in above a threshold, and disputes are routed and summarised with AI assistance before a human decides.
naxcrow.com
Laravel · React · Flutter · payment APIs
The constraint
Buyers and sellers across African markets who have no reason to trust each other, and money that has to sit somewhere neutral until they do.
What we built
Buyer and seller funds held against milestones, payments clearing through several gateways, KYC stepping in above a threshold, disputes routed and summarised with AI assistance before a human decides, and Flutter apps with push on both stores.
What it taught us
The dispute path is the product. Everything else is the happy case, and the happy case is the easy half.
A custom built SaaS for the US nursing market, delivered for an American client. RN, LPN, CNA and advanced practice roles across the country, employer listings and plans, search by type, category, location and salary, and the admin tooling behind it all.
A jobs board lives or dies on search and on trust. Nurses need to find the right shift in the right city fast, and hospitals need to believe their listing is reaching real candidates before they pay.
What we built
A custom SaaS from the ground up, not a plugin. Search across job type, category, location and salary, employer profiles with live listings, four employer plans behind one entitlement model, a featured employers tier, and the admin surface that keeps the listings honest.
What it taught us
The founder came to software from nursing. Building the thing she once would have used is a different kind of specification.
A research toolkit for a market structure model: parallelised backtesting, walk forward validation, a locked case regression harness, and a zero build chart viewer that draws every structural read bar for bar so a human can check it by eye.
github.com · public repository, and this is the real interface
Python · pandas · pyarrow · walk forward
The constraint
A backtest that quietly peeks at future data will show you an edge that does not exist, and you will not find out until it is live.
What we built
A parallelised backtesting and walk forward validation engine, a locked case regression harness that fails the build when results drift, and a zero build web viewer that draws the model's structural reads bar for bar so a human can check them by eye.
What it taught us
Being able to see what the engine saw is worth more than another statistic about it.
A multi broker futures trading system that takes a strategy from research through validated backtest to live unattended execution. Live market data, a twelve check risk manager, four broker bridges behind one interface, and a causal simulator built so look ahead bias is impossible by construction.
github.com · public repository, and this is the real interface
Python · asyncio · FastAPI · ZeroMQ · PostgreSQL
The constraint
A strategy is only worth deploying if the backtest was honest, and only survivable if execution keeps running unattended when a broker, a feed or a server drops.
What we built
A two node system: orchestration, data feed, dashboard and database on one server, broker execution on another, talking over ZeroMQ. Live futures data with gap detection and back fill, a twelve check risk manager plus a drawdown governor, four broker bridges behind one interface, breakeven and trailing logic, a reconciler that rebuilds live state after any restart, and an operations dashboard with a one click flatten and pause.
What it taught us
The simulator is the product. Strategies receive only a read only buffer of past bars, which makes look ahead structurally impossible rather than merely discouraged.
A security provider that needed its site to argue competence rather than assert it. Security first architecture, performance hardening, a custom live chat widget that captures enquiries when nobody is online, and service pages built to convert.
A United Kingdom cybersecurity provider whose own site had to argue security competence rather than assert it.
What we built
Security first architecture, performance hardening, a custom live chat widget that captures enquiries when nobody is online, and service pages built to convert.
What it taught us
For a security firm the site is the first audit. Anything sloppy in the front end is read as sloppiness in the product.
