Insights
Practical lessons from putting AI, databases and cloud infrastructure into production, drawn from the systems we build and run for our clients and our own products.
-
Setting AI Confidence Thresholds from the Cost of an Error: Lessons from Cerberus ID
How Cerberus ID turns a fraud score into approve, check and reject zones, and how calibrating against 10,806 real registrations raised auto-approval precision to 99.5% with 78% fewer false approvals.
-
How Our Spam Filter Decides Over 78% of Messages Before Calling an LLM
SPAM Escudo filters contact-form spam in five stages, with the LLM last. Every LLM verdict teaches the cheap stages to catch more, and a weekly AI-assisted review of what still reaches the…
-
Local AI on a CPU: How a 270M-Parameter Model Makes Invoice Processing Practical
How Transit Ledger processes invoices and receipts entirely on a local CPU server: a tiny classifier gates a 4B extraction model, code checks the totals, and real-world data exposed a shortcut the…
-
AI Agents in Production: From Experimentation to Business Value
How to take AI from experiment to production while keeping cost, accuracy and risk under control: a cascade that uses the right technology for each step, with real results from spam filtering,…