Plates / 003 · Polymarket bot + Tradofire

Plate III · 003 2023 — present Lisbon · Independent Algorithmic trading

Polymarket bot + Tradofire.

Two real‑money trading systems running in production — one against prediction markets, one across multiple crypto exchanges.

FIG. III.A · Frontispiece Live position curve
+ orderbook
· 30‑day window ·

Fig. III.A Live positions across a thirty‑day window. Real money. Detail withheld in this public edition.

§ 01The problem

Markets that don’t wait.

Two questions I wanted to answer for myself: could I build a profitable algorithmic system end‑to‑end and keep it alive in production? And could I do it as the only engineer responsible — for the strategy, the execution path, the risk controls, the failure modes, all of it?

The Polymarket bot trades a portfolio of strategies across prediction markets. Tradofire is the broader crypto system — multi‑exchange data ingestion, an event‑driven execution engine, and the kind of unglamorous risk and reconciliation code that lets you sleep through a flash crash.

Real‑money systems are unambiguous. They’re either making money or they aren’t, and the codebase will tell you which.

§ 02My role

Sole engineer, founder.

RoleSole engineer & founder

Tenure2023 — present

CapitalReal money in production

Reports toMyself — expensively

End‑to‑end ownership in the most literal sense: strategy research, code, infrastructure, monitoring, the on‑call rotation, the post‑mortems.

Polymarket. Live, profitable bot trading several strategies in prediction markets. Automated execution, market‑data handling, position management — and the loud alerts when any of those drift away from where they should be.

Tradofire. Multi‑exchange crypto platform ingesting WebSocket market data from Binance and Coinbase, with an event‑driven execution engine, secure API integrations and risk controls. Designed for low‑latency operation and for the failure modes you only learn about by being on the wrong side of one.

§ 03The stack

What it’s built on.

StrategyStatistical and event‑based strategies; quick backtesting harness. Python · pandas · NumPy

Market dataReal‑time WebSocket ingestion from Polymarket, Binance, Coinbase. WebSocket · async I/O

ExecutionEvent‑driven order engine with per‑strategy adapters. Python · asyncio · Postgres

RiskPosition limits, drawdown caps, kill‑switch wired into every adapter. Risk engine · alerts · circuit breakers

InfrastructureAlways‑on services, monitoring, secure key handling. Docker · systemd · Vault · Prometheus

FIG. III.B Strategy PnL curve
· thirty‑day rolling ·

FIG. III.C Risk‑control
state machine
· kill‑switch path ·

Fig. III.B–C A strategy’s PnL curve and the risk‑control state machine that wraps every order path.

§ 04Outcomes

What runs.

01Live. Polymarket bot in production, profitable to date.
02Tradofire alpha. Multi‑exchange platform running against Binance and Coinbase with risk controls in place.
03Operational lessons. Three real outages, three post‑mortems, three structural fixes — the kind of evidence the day job doesn’t give you.
04Capital allocation. Live capital allocated, not paper‑traded.

Detailed P&L and strategy specifics are not in the public edition. Available under NDA.

Endnote

You learn very different things about a system when the failure mode costs you money.

— from a 2024 post‑mortem.