11 areas
Eleven areas, and behind each one a specific system that is running right now. If an area is not on the list, it does not mean we will not take it on: it means we have no proof to show, and we will say so plainly.
High-load transactional systems
Wallets, payments, accruals, bets — everything where an error in a single line of code turns into a real loss. We build them so that a payment cannot go through twice and a balance cannot fall below zero, even when thousands of people pay in the very same second. And if an operation gets stuck somewhere between your system and the bank, it will not hang there: the system either drives it to completion or returns the money.
The platform we look after handles about 1.5M money operations per day. After we rewrote payout settlement, 267 double payouts disappeared along with a queue of 437 thousand stuck jobs.
Payment integration and fintech
We connect your product to whatever your customers actually pay with: bank transfers, mobile money, cryptocurrency, cash over a counter or through a terminal. Every payment provider speaks its own way, and half of them speak a dialect a decade old. For us that is routine. If the market has no suitable gateway, we build our own — ours already has 21 payment systems connected.
More than 20 integrations across the Caribbean, Latin America and the CIS. Plus our own money transfer network: a person hands in cash in one country and the recipient collects it in another with a PIN code.
Crypto processing
A client pays you in USDT, bitcoin or another coin — you receive the amount in local currency and never think about the blockchain. The rate is picked automatically from real orders on exchanges, incoming payments are screened for dirty origin, and wallet keys are stored encrypted and decrypted only at the moment of the transfer.
Our gateway works with 10 networks — Ethereum, TRON, Solana, TON, Bitcoin and others. More than 85M operations have gone through it, and the database has grown to 283 GB.
Real-time systems and streaming data
Betting odds, market quotes, sensor readings — everything that changes several times a second and has to reach thousands of screens at once. Simply moving the data is not enough here: the numbers must not jump in front of a person, nothing may be lost when the connection drops, and events must not swap places. We wrote such a core from scratch and know where it breaks.
Our core takes a quote stream across 12 sports and distributes it to thousands of browsers at once. It has been in production since 2020.
Live video and remote equipment control
We show hundreds and thousands of live video streams in a browser — no plugins, with a delay of fractions of a second. Our most vivid project: a person plays a real slot machine from a browser while the machine physically stands in a hall. A tap on the screen reaches the board inside the cabinet and closes a contact, the machine responds, the camera shows the result. The same approach works for any equipment that has to be operated at a distance with the result visible to the eye.
About 1000 connected machines and roughly 2000 simultaneous video streams: a screen and a camera for every cabinet.
Embedded systems and electronics
We make the hardware end to end instead of assembling a project out of parts made by others: we draw the schematic and the printed board, write the firmware, raise the server the device reports to, and the interface where a person sees that data. Separately we write drivers for industrial machinery — bill acceptors, cash dispensers, receipt printers, card readers.
Our own boards and firmware on ESP32, STM32 and AVR — including a network board for which we had to write our own network stack: the off-the-shelf one did not fit into the microcontroller memory.
Data and analytics
Reports that do not take minutes to answer and do not drag the production database down. Data streams out of the working system into a separate store built to count across billions of rows, and the report is built there, without getting in the way of people paying and placing orders. The full history stays, not just the last ninety days.
Reports that used to hit the 15-second limit of the database now take 0.2–10.6 seconds. Exports and paging work over a 90-million-row table without timeouts. The hot database itself was compressed 45-fold, without stopping payments.
iGaming: platforms, games, mathematics
A gaming platform in full: the site for players, the cash desk in physical outlets, the back office, the affiliate programme, integration of external game providers, a sportsbook module and lotteries. We do the game mathematics ourselves: the return percentage is configurable and verified by calculation, not by eye.
12 in-house games, 8 connected external providers, 159 slot mathematics cores with a configurable return percentage and built-in Monte Carlo verification. The lottery platform runs on official US draws.
Mobile development
Applications for Android and iOS: both the ordinary kind from the store and the kind that lives outside it, updating itself where the store is unavailable or does not fit the rules. We also build the unusual, such as an application that turns a phone into an SMS gateway where the usual operator services do not work.
Android applications that update themselves bypassing the stores, native iOS and Android, a Unity client with hardware video decoding, and phone gateways carrying about 7.5 thousand SMS a week.
QA and performance engineering
Not clicking through before a release, but finding where the system hits its ceiling and removing that ceiling. We write our own load generators, reproduce the peak hour before it happens, and reconcile the money down to the cent: how much came in, how much was credited, whether it balances.
Our load generator with 274 virtual players exposed the bottleneck in bet settlement: after the rework the rate grew from 290 to 4360 bets per minute. The financial audit of the runs across 123 543 bets found zero double payouts and zero lost credits.
DevOps and operations
We take on what has to run around the clock: servers, releases, monitoring, backups and being on call when something breaks. Updates roll out automatically and identically instead of by hand over SSH, and several independent monitoring tracks watch the system so that one of them failing does not leave you blind.
An in-house build factory across 301 projects and 15+ runner hosts, 47 roles for automated server setup, five independent monitoring tracks, backups with mandatory restore verification, and a post-mortem for every serious incident.