Teams & systems that
hold up under load.
The systems your business runs on — and the team that runs them.
Twelve years, from first commit to steady state.
How an engagement unfolds,
start to finish.
From setting direction and hiring, through building the systems, to securing and running them in production.
Fractional tech lead / advisory
define goals, set strategy
Hiring & team building
interview loops, bar-raising
Building from the ground up
0→1 greenfield systems
Backend architecture & microservices
event-driven, DDD
Legacy → cloud migrations
monolith → EKS / Terraform
Engineering processes & velocity
CI/CD, faster TTM
Performance & scalability audits
high-load, latency
Something else?
let's scope it
Security & compliance
SOC2, data protection
FinOps & cost optimization
right-sizing, spot, budgets
Case studies.
A few systems I've shipped.
Terraform on AWS — the watchtower: cloud state, locks, and the deploy role
One bootstrap stack owns state, locks and deploy roles — no collisions, no standing admin keys.
DevOps for a SOC 2 audit — Part 1: Docker as an evidence factory
Every control a build stage, every result an exportable artifact for CI and the analyzers.
A temporal EAV entity store in Postgres — Part 1: Storage concept
One schema, a dozen domain types — versioned fields, point-in-time reads, zero migrations.
A temporal EAV entity store in Postgres — Part 2: The abstract repository
One abstract class turns the storage into a complete CRUD interface per domain type.
A temporal EAV entity store in Postgres — Part 3: Services and the API edge
Business rules, cross-entity transactions, and time-travel exposed as a query parameter.
A temporal EAV entity store in Postgres — Part 4: E2E testing
The suite as the executable specification — three assertion altitudes against real Postgres.
A temporal EAV entity store in Postgres — Part 5: Maintenance and migration
Invariant queries on patrol, Prometheus alerts, and entity migrations for a schema with no DDL.
Notes on building systems.
All writing →Closing the gap to llama.cpp on a desk-side GB10 — six Atlas iterations, measured
Serving Qwen3.8-Flash-Next on the Atlas engine: per iteration the analysis, the decisions, the tower-defense benchmark and the defects — 15 → 21.6 tok/s against llama.cpp's 24–25, every engine change proven bit-exact.
Benchmarking agentic flows on a desk-side GB10 — from model funnel to a shipped PR
Three research rounds over real codebases, verification-first grading, an orchestrated harness — and four fixes landing upstream on the Atlas inference engine.
Benchmarking local models by making them ship — a tower-defense build-off
Prod 27B vs Qwen3.8-Flash-Next quants: one complete HTML5 game each from identical prompts — speed, prefill, craft, and the playable results.
A temporal EAV entity store in Postgres — Part 5: Maintenance and migration
pg_stat metrics, invariant queries on patrol, Prometheus alerts, and migrations without DDL.
A temporal EAV entity store in Postgres — Part 4: E2E testing
Three assertion altitudes, the behavior matrix as a test plan, rollbacks proven at the row level.
A temporal EAV entity store in Postgres — Part 3: Services and the API edge
Cross-entity transactions, the validation ladder, and time-travel as a query parameter.
Twelve years across billing systems and B2B SaaS — Rust and TypeScript on the backend, AWS and Kubernetes underneath. I've led migrations off legacy monoliths, built teams from the first hire, and kept enterprise SLAs intact while cutting cloud spend.
Have a system that's
too important to break?
Tell me what you're building. I reply within a day.