I build AI systems that do real work — AML detection, blockchain investigation, reasoning engines. These aren't demos. Here's what they've produced.
28 detection rules running against real Ethereum transaction data. Catches mixer activity, sanctions violations, structuring patterns, and wallet risk profiling.
Paste any Ethereum address, get a full risk breakdown. The engine runs 28 rules against live chain data and generates an AI-powered risk assessment.
What happens when you make an LLM audit its own reasoning? A year of building a cross-model symbolic verification system. Four rounds of adversarial testing.
Multiple specialized AI agents, each doing one job well. Detection agents scan blockchain data. Reasoning agents cross-verify outputs. No single model trusts its own answer — everything gets checked against independent sources.
Pattern matching against known exploit signatures, sanctions lists, mixer activity, and behavioral anomalies.
No agent validates its own output. Every critical finding gets cross-checked against external ground truth — OFAC lists, DeFiLlama, public chain data.
Symbolic reasoning over raw outputs. Not just "what happened" but "why it matters" — connecting transaction patterns to known threat models.
Everyone is selling you on building an agent. Nobody is telling you what breaks when they run at 3am and you are asleep. I have tried more than a dozen. A handful are still breathing. The rest taught me rules I now enforce automatically. Here is the honest field report, and a direct ask at the end to anyone else quietly doing this work.
Real numbers from 2025 and 2026. 55,000 AI-attributed layoffs. 44 percent of finance teams running agentic AI. 1.3 fewer hours worked per week in the Altman UBI study. The compression curve is not a forecast anymore. It is a measurement. Here is how to read it.
Real production data from JPMorgan, HDFC, ANZ, RBC, Scotiabank, and CIBC. 99% of banks plan autonomous agents. Only 11% have deployed. Where does Canada stand?
88% of organizations use AI. Only 21% have agentic AI in production. The space between adoption and deployment is where the next decade of banking gets built.
A data-backed look at how Canada's biggest banks are adopting AI, from RBC's 10-year head start to CIBC's CAI rollout that saved 600,000 hours.
Cisco scanned 31,000 OpenClaw skills. 26% were malicious. Here's what's actually happening with AI agent security — real CVEs, real case studies, and what defense looks like.