Technical writing from our team. We cover what we are building, what worked, and what did not.
Our LSTM models looked great on backtest but degraded fast in live trading. Here is why XGBoost with hand-crafted features outperformed deep learning for our ASX signal pipeline.
We tested 14 sentiment features derived from Reddit, X, and on-chain data. Only 3 had statistically significant predictive power after transaction costs. Here are the results.
Markets change regimes. Our models retrain nightly, but that is not enough. We built a drift detection system that triggers emergency retraining when feature distributions shift beyond thresholds.
We built Echo (iOS + Android) in 10 weeks with a team of three. Here is what we got right, what we got wrong, and what we would do differently next time.
A walkthrough of our streaming architecture: Kafka topics per exchange, Avro schemas, exactly-once semantics, and the backfill system that saved us during the March outage.
The 23 features that actually matter in our equity signal models. Includes rolling volatility ratios, order flow imbalance, and why we stopped using RSI.
We apply the same technical rigour to client projects. Tell us what you are building.