Resources
Technical content from the team that built Compiled.
Research, architecture deep-dives, and perspective from the people who built the math.
The AUC 0.995 Insider Trading Detector
How we built a production-quality detector using only synthetic training data. End-to-end: synthetic generation, contrastive training, hard-negative validation.
Zero False Positives: Real-World Validation
Testing synthetic-trained detectors blind against real SEC/DOJ enforcement communications vs real Federal Reserve/earnings communications. What we found.
Federated Threat Intelligence Without Sharing Data
Differential privacy, centroid geometry, and the math behind AUC 0.74→0.86. How the network effect compounds without moving a single message.
The Policy Compiler: From English to Mathematics
Architecture of the Compile step: embedding, synthetic generation, linear classifier training, hard-negative testing, and artifact packaging.
AI Agent Governance: The 2025 Landscape
Where current frameworks fail and what mathematical governance enables. Why prompts and rules can't keep up with agents operating at machine speed.
Communications Surveillance in the Age of AI Agents
When your agents generate communications, your surveillance system needs to adapt. The new threat surface and how semantic detection addresses it.
Stay current.
We publish technical content when we have something worth saying. No filler.
Or follow the research: security@compiled.ai