Notes from the field.
On distributed systems, the maths underneath them, and the leadership work that turns design into shipped software.
This blog is written from production, not from documentation: AI decisioning systems in regulated businesses, agent architectures that survive audits, queueing and capacity maths, and the trade-offs behind cloud and event-driven designs. Every post is anchored to a system I have built or reviewed, with the numbers I can defend. Longer pieces are cross-published on Coding Layman, where the same ideas are rewritten in plain language for readers earlier in the journey. If you want the short version of any article, each one opens with a TL;DR built for skimming.
Three things my agent guard could not see
My coordinator stopped parallel coding agents colliding. Then I read what it missed: a check grading a fiction, a write to an unseen repo, a dead lease. · 7 min readTwo agents, two repositories, one refused collision
Two coding agents ran in parallel under declared write scope. One collision was refused before spawn, one scope escape was caught at exit. · 6 min readApprove and Deny Are the Easy Part: Loan Decisioning With a Human in the Loop
Approve and deny are the easy majority of loan volume. The architecture lives in the exceptions: classify, route, human review, re-entry, audit. · 12 min readProduction RAG for a Tier-1 Bank: An Architecture Walkthrough
Bank RAG projects die at second-line risk review, not at the model. The retrieval architecture that survives an audit, and why it was chosen. · 17 min readEasy to Demo, Hard to Operate: Why AI Pilots Stall in Month Two
AI pilots rarely fail on the model. They stall on memory, security, monitoring and cost. One platform built inside hard limits shows what that takes. · 9 min readYour coding agents do not need better merges, they need to declare
Isolation then merge checks for collisions after the work is paid for. Claim-before-spawn checks before the process starts, so refusing costs nothing. · 6 min readI built a supervisor for my coding agents, measured it, and deleted it
I built a supervisor loop for my coding agents, measured it against no supervisor, and removed it: identical success, two percent worse cost. · 6 min readThe AI Loan Officer Exists. The Hard Part Is Everything Around It.
AI loan decisions are lifecycles, not responses: async intake, queued workers, human-review loops, and an audit trail regulators can trust - end to end. · 7 min readThe Agent Trio: How I Run Three AI Agents on One Shared Memory
Three agent planes, one shared memory, zero self-modification: the shape that keeps a multi-agent system debuggable as it grows. · 6 min readWhy Architects Can't Ignore Knowledge Distillation of Black-Box LLMs
Proprietary LLM costs explode at scale. Here's how to compress frontier models into deployable, cost-effective alternatives you actually control. · 8 min readMasking Sensitive IDs Across 600 Million Documents: A Batch Architecture That Scales
How to design a production batch pipeline to mask the first 8 digits of an ID number across 600 million documents, focused on processing time, cost, components, and compliance. · 6 min readWhy Numbers Are Not Enough: Scalars vs Vectors
Most people believe mathematics starts with numbers. That belief works-until it doesn't. · 5 min readToken-Based vs Session-Based Security: An Architect's Perspective
A practical breakdown of session vs token security, when to use which, and modern best practices. · 8 min readThe Insight That Unlocks Linear Algebra for Engineers
Discover why matrices are not just tables of numbers, but functions in disguise. A paradigm shift that transforms how engineers think about systems, transformations, and architecture. · 6 min readNew essays, straight to your inbox.
Occasional, in-depth writing on distributed systems, AI agent architecture, and engineering leadership. No spam, unsubscribe anytime.