High-Scale Production Systems
Real metrics from systems running in production across Tata Group, Globe Telecom, and the AFL.
#1How I Cut Tata Group's BigQuery Bill by 57%
57% data warehouse cost reduction. Query refactoring, MERGE redesign, capacity-based slot reservation.
#2
Globe — running a 30K+ TPS transaction platform on Kubernetes
Distributed transaction engine for Globe Telecom Philippines. Kafka-backed event sourcing, Redis-based idempotency, exactly-once processing guarantees.
#3
Brownlow — zero-trust voting on Cloud Run during live AFL broadcasts
100K+ votes, 10K+ concurrent users during live broadcast. Zero-trust architecture with Cloud KMS for vote integrity and GraphQL federation.
#4
The MERGE pattern that cost ten times more than INSERT-then-UPDATE
Single biggest FinOps lever in the Tata engagement. How a common SQL pattern silently 10x'd costs and what replaced it.
Multi-Agent AI & Architecture
Designing, migrating, and operating multi-agent systems on the Microsoft Agent Framework in production.
#5Why We Migrated from Google ADK to the Microsoft Agent Framework
The philosophy, trade-offs, and what we learned converting 18+ agents in 3 months. Provider abstraction as the foundation for portable agents.
#6
Lessons from Converting 18 Agents in 90 Days
What worked, what was hard, and what we'd do differently. Real numbers: 18 agents, 90 days, 5 governance policies, 4 provider swaps.
#7
Agentic architecture on the Microsoft Agent Framework — the seven load-bearing pieces
Protocol, registry, event bus, governance engine, orchestrator, observability, evaluation. The seven components every multi-agent system needs.
#8
OpenTelemetry and Evaluation in Multi-Agent Workflows
The full production observability stack for multi-agent systems. Everything learned building Genie — traces, metrics, evaluation pipelines.
#9
Why Go for production agentic AI
Why boring tech wins in regulated environments. Principled argument for Go over Python in production multi-agent systems.
Security Architecture
Production security across RBAC, database-level isolation, token exchange, and operations playbooks.
#10Consolidated security deep-dive — RBAC, RLS, token exchange, encryption, audit
Long-form security narrative covering auth, authorization, tenant isolation, envelope encryption, append-only audit, and governance enforcement.
#11
Agentic security in production — the operations playbook
12 months of production multi-agent security. SLIs, incident runbooks, drift detection, adversarial testing in a regulated environment.
#12
Mapping a multi-agent platform to the GCP PCSE blueprint
Every PCSE exam domain mapped to production code in Genie. Cloud security certification knowledge applied to real multi-agent architecture.
#13
PostgreSQL Row-Level Security Is HIPAA Defense in Depth
Database-level enforcement, not app-level. Why RLS policies at the PostgreSQL layer are a stronger HIPAA control than application middleware.
Healthcare AI & Compliance
HIPAA, Cures Act, FHIR, and diagnostic AI — where regulation meets production code.
#14Building a HIPAA-Aware Medical AI Platform in Go
What HIPAA looks like when you express it as Go interfaces — governance policies, append-only audit, PHI redaction at the logger seam.
#15
"Enough to Reconstruct, Never Enough to Leak": The HIPAA Audit Log Design Problem
The hard audit problem: log enough to satisfy regulators, never enough to leak PHI. Solving § 164.312(b) at the design level.
#16
The 21st Century Cures Act, Expressed in Go
Translating §3060 Clinical Decision Support requirements into working Go code. Regulation as executable policy.
#17
Moving Diagnostic Accuracy 42.9% → 85.7% by Changing Two Files
Doubling diagnostic accuracy with principled evaluation, not model changes. Prompt engineering and evaluation methodology in medical AI.
Regulatory Governance & Data Infrastructure
RBI FREE-AI compliance, AI governance as code, and zero-downtime database migration.
#18RBI FREE-AI implementation notes — 26 recommendations to file paths
All 26 RBI FREE-AI recommendations mapped to working code. Board-level policy expressed as YAML and enforced at runtime.
#19
AI governance — from credential to codebase
Board policy as YAML, governance as code. Bridging business requirements and engineering with enforceable AI governance.
#20
CDC for minimal-downtime Spanner migration — Datastream + Pub/Sub + Dataflow
Zero-downtime migration patterns with Change Data Capture. Datastream → Pub/Sub → Dataflow pipeline for PostgreSQL to Cloud Spanner.
Governed Multi-Agent Patterns
Patterns from building a governed multi-agent system on the Microsoft Agent Framework — policy, identity, and reliability where the model is untrusted input.
#21Policy Is Code, Not a Prompt
A gateway that returns FORBID for the delete tool can't be talked out of it. Put enforcement where the model can't reach it and prompt injection stops mattering for access control.
#22
Give Every Agent Its Own Credential
One workload identity means one hijacked agent has every agent's blast radius. Split identity across app, cloud, and crypto layers to shrink it to a single role.
#23
"maximum recursion depth exceeded" — a durable-workflow checkpointing war story
Threading an SDK client through a checkpointed workflow crashed on circular references. The checkpoint boundary is a serialization boundary: JSON only, values and keys.
#24
EXHAUSTED Is Not Failure: Bounding Agent Runs
Cap every run at max_turns and make EXHAUSTED a first-class outcome — not a failure, not a silent success — so a runaway loop becomes routine and observable.
#25
The Cheapest Reliable Executor Wins
Deterministic rules get first refusal at zero cost and still fire when the model is down; only the novel long tail escalates to graduated agents. Build a ladder, not a model call.
These 25 articles are selected from 340+ published posts.