Skip to content
Menu

I build AI systems that have to work.

Agents, evaluations and production infrastructure, engineered with evidence, validation and rollback. Principal and forward-deployed AI engineering, on contract or retainer, fully remote.

Tell me what you're buildingSee it in production

Taking contract and retainer engagements

varai.com: The operating system for agentic commerceVARAi · AI commerce for eyewearI’m CTO. Live in three regions, ~150 stores live or onboarding.
  • VARAiCTO & forward-deployed AI engineer
  • Netflixbought the system we built, 2018–2020
  • Global optics & lens enterpriseNDA · ten apps
  • Disneystudio security approval at Arch
  • Arch Platformlead developer, 2016–2021
  1. E-01~150 storesLive or onboarding on an AI commerce platform I run as CTOVARAi: AI search, 3D design, a deterministic lens engine, three regionsVARAi, 2026
  2. E-0221 → 0 lostEnterprise tenants cut over live between regions, zero data lossVARAi, EU → India, with row-count safety gatesVARAi, 2026
  3. E-03219 auditedArchitecture assurance for AI-built code: decisions checked against the code, every verdict evidence-backedAgentOps, my audit harness, on a large production codebaserun 2026-09-28
  4. E-04−50%Cloud cost on a studio VFX pipeline; Aspera-speed sync without AsperaNetflix artists on Arch Platform, Vancouver ↔ LA in sync nightly2018 – 2020

Selected work

All case files
C-05 2025 – nowLive

VARAi: the operating system for agentic commerce

VARAi · eyewear, from design to sale CTO & forward-deployed AI engineer

An AI commerce platform for eyewear, live in three regions with ~150 stores live or onboarding. I run its engineering, and I built the AI search, the AI 3D designer, the deterministic lens engine and the agentic delivery pipeline.

stores live or onboarding
~150
tests on the prescription-lens rules engine
1,500+
enterprise tenants cut over live, EU → India
21 → 0 lost

Gemini + Claude · Redis · three.js · Magento + Shopify · GCP, 3 regions · Claude Code

Read the case file
C-07 2013 – nowShipped

Enterprise cloud engineering: Netflix, Arch & Atlascloud

Netflix (via Arch Platform) · Arch Platform · Atlascloud Cloud architect

Kept Netflix artists in Vancouver and LA working on one show, with terabytes synced every night. Alongside it, a cloud VFX studio that passed Disney's security review. Underneath, years of building the hardware.

cloud cost, idle instances shut down nightly
−50%
studio throughput, no new hardware
3×
Netflix took the sync system in-house
bought

AWS · Terraform · Python · rsync · KVM / Proxmox

Read the case file
C-01 2026Live

AgentOps: AI agents you can audit

My own tool, run on a production codebase Sole author

Checks whether 200+ architecture decisions still describe the code. A deterministic gatherer freezes the evidence, one schema-bound model call judges it, and a validator, not the model, decides what counts.

every accepted verdict traced to frozen evidence
evidence-backed
accepted / rejected by the validator, 2026-09-28 run of 219
146 / 37
byte-identical rebuilds, one-command rollback
reproducible

Python · OpenAI + Gemini · JSON Schema · SQLite · GCP

Read the case file

How I engineer: AI agents you can audit.

Deterministic evidence collection, one bounded model call, a validator that decides what counts, and reproducible runs. It’s the discipline I bring to every system, and AgentOps is where I built it in full.

Read the AgentOps case
AgentOps · first corpus run2026-09-28
run
run-20260928T172120Z
ADRs targeted
219
dispatched
186
provider rate limit stopped the run; 24 pending
validated
146
71 drifted · 57 current · 18 status wrong
rejected by validator
37
model calls / tokens
336 / 9.7M
accuracy
not published
pre-registered study, independent judge, in progress
release suite
802 / 155
assertions / guarantees, all green

rebuild: byte-identical rollback: one command

146 validated

A real receipt from AgentOps, my audit harness: every accepted verdict traced to frozen evidence, and no accuracy figure until an independent judge has scored a pre-registered sample.

How I can help

Engagement details

You don't hire me to experiment with AI. You hire me when the experiment needs to become engineering: from a fixed-scope audit to building and launching the production system.

  1. S-01Production AI Systems / Forward DeploymentFor companies with an AI initiative that needs to become a production system. I work from architecture through implementation and launch: model integration, APIs, data, infrastructure, evaluation and deployment.
  2. S-02Architecture Drift AuditFor teams whose architecture decision records, design docs or runbooks no longer match the code, often after a year of fast shipping or heavy use of AI coding agents. Call me before a migration, an acquisition review or a compliance audit.
  3. S-03Agent Reliability / Harness BuildFor teams with agents that work in the demo and wobble in production. I build the harness around them so every run is evidenced, checked and reversible, and a wrong answer is caught before it reaches a user.
  4. S-04AI Evaluation DesignFor teams that can't say whether the last prompt, model or fine-tune made things better. I design evals that answer that question honestly, including ones that are hard to game.
  5. S-05Fractional AI LeadFor founders and engineering leaders who need senior AI judgment and hands-on delivery without a full-time hire. I set the architecture and the agent-team process, then write code alongside your engineers.

Not a fit for small-business automation or no-code workflow setups. Senior contract, retainer and fixed-term leadership engagements only.

Physical AI & robotics

AI doesn't stop at software. I'm bringing production AI engineering into robots and physical systems: choosing the hardware, integrating its SDK, and building the software, agents and operations layer around it.

Explore robotics
  • Hardware selection
  • SDK integration
  • AI agents
  • Fleet software
  • Custom automation
  • Deployment

from a collaborator

“Bron is one of the best last-mile devs I know.”
Brad Ross · AI Agentics Engineer

Bring me the part that has to work.

I don't stop at recommendations. I build the thing the recommendation calls for. Send me the problem, the stack and the deadline, and I'll tell you plainly whether I'm the right person for it.

Contract CTO · Principal AI engineering · Forward-deployed AI engineering · Enterprise AI architecture

Start a conversationor message me on LinkedIn