Writing a tilt detector that ran in production for 6 months without a false positive
A deep-dive into the feature engineering, HMM architecture, and real-time inference pipeline behind TradeVision's behavioural alerts.
Architecture, trade-offs, post-mortems, and the occasional deep-dive. Written by the people who ship the product.
A deep-dive into the feature engineering, HMM architecture, and real-time inference pipeline behind TradeVision's behavioural alerts.
We pick boring, battle-tested technology over hype. Postgres before Mongo. React before Svelte. Our users want reliability, not novelty.
Every 100ms matters when you're making a decision. Sub-50ms p99 isn't a nice-to-have; it's the product.
We don't read your trades. PII-free telemetry means we lose some signal — and keep your trust.
We average 6+ deploys per day. Post-mortems are public when they affect users. Learning is the point.
Code is read 10× more than it's written. Every PR gets a description. Every non-trivial function gets a docstring.
We're hiring across infra, ML, and full-stack. Remote-first. Always small, never bureaucratic.