The part everyone skips.
How I built a trading system that runs on Hyperliquid 24/7 — real capital, no human watching charts. What I got wrong three times before getting it right.
The WhatsApp bot was the first thing I built that felt autonomous. Not automated — autonomous. What that difference actually means when you're building systems
Every company that goes bankrupt files a public petition. PACER publishes those filings 2-3 weeks before anything reaches the liquidation platforms. I built a m
Migrating the entire system from Anthropic to MiniMax-M2.7 in one day. 2,589 tests passing, zero failures. Here's what the cleanup looked like.
Angelfish is my stat-arb trading system — pairs trading on correlated crypto assets. How I designed it, what the backtests showed, and what I changed before goi
I built a bot that monitors Amazon and high street retailers for price gaps. What I found when I ran it — and what the data said about whether retail-to-Amazon
I built a mentor system that watches my decisions and pushes back. The goal wasn't to have an AI agree with me — it was to have one that doesn't.
Walk-forward validation, hypothesis testing, mutation testing. How I built a testing framework that makes it hard to fool myself about whether a strategy has ed
I ran Archer (Claude reasoning) and Sniper (pure Python signals) simultaneously on Discord, same capital split, to see which made better decisions.
I built Archer — a trading agent where Claude makes go/no-go decisions on pre-filtered signals. Python enforces the risk. Claude does the reasoning.
I used a 60-qubit quantum processor to generate the random numbers for my trading system's stop-loss jitter. The reason is less strange than it sounds.
A three-factor synthetic binary scalper with hexagonal architecture. Orderbook imbalance, momentum, and funding rate — combined into one score.
Four Hyperliquid API bugs in one week. Rate limit storms, orphaned positions, an AVAX dust position that wouldn't close. How I found and fixed them.
My Discord bot answered questions using whatever was in its context window. I wanted it to remember. So I built a RAG system with ChromaDB and wired it in.
Most trading circuit breakers are arbitrary. Mine uses binomial distribution to calculate the exact number of consecutive losses that shouldn't happen by chance
ViralOS runs 15 Instagram content concepts at once, scores them every 24 hours, kills losers, and redeploys winners. Here's what that looks like under the hood.
An AI-powered budget vs actual forecasting service — Xero integration, GL code learning, citation trails for every number. Built for a real client.
I built a data truth gate for my trading engine after realising backtest results could be faked by bad data. What I found — and the invariant that fixed it.
The first commit was December 9, 2025. One line in the message: REAL QuantDesk V4 — FULL PROJECT UPLOAD. Here's what that meant.