FRACTIONAL PRINCIPAL ARCHITECT · every claim on this page is real
For founders, CTOs, and engineering leaders shipping B2B SaaS, payments, or AI-assisted products.
AI made shipping code cheap. It didn't make shipping products safe. I'm the architect teams bring in a few days a week so that velocity doesn't turn into breaches, outages, and cloud bills.
Your team ships faster than ever. Your security, architecture, and operational discipline don't.
The failures land later - as incidents, licensing problems, and invoices. My job is to make sure they don't land at all.
Your senior-most technical voice, part-time.
Monthly retainer, sized by days/week
Fits: teams of 2-20 engineers shipping fast without a principal-level architect on staff.
Relevant evidence: security flaws stopped before launch · cloud spend cut 84% →A structured audit of what you're about to ship.
Fixed price, quoted on repo count & scope
Fits: pre-launch, pre-fundraise, or post-incident.
Relevant evidence: a payment release stopped before every refund could fail →How your team actually uses AI coding tools - and what that's exposing.
Fixed price, one week, one deliverable
Fits: any team where AI writes a meaningful share of the code.
Relevant evidence: AI-agent policy created after a real credential leak →A 30-minute risk review gives us the product, the release, and the decision that matters most.
Choose the smallest engagement that resolves the highest-risk unknown.
Prioritized findings, a decision log, and a clear next move stay with your team.
Caught in a pre-implementation review, not an incident. Stopped feature work, audited the entire platform for the same class of flaw, closed the process gap that allowed it - then fixed the code.
~$10K/year saved at pilot scale. Wrote the non-negotiable invariants first: no monolith collapse, every service boundary preserved, and a documented one-weekend path back to Kubernetes.
A live end-to-end test showed the provider contract differed from what the docs implied. Declared NO-GO on my own evidence, fixed it, and proved exactly-once refund semantics under replay before shipping.
An AI coding agent leaked credentials into a session log. I wrote a machine-followable secret-handling and supply-chain policy, propagated it into every agent session, and wired an automated leak check into the deploy pipeline.
Six review agents fanned out across a seven-repo feature. 48 raw candidates, each confirmed by direct code read; one killed as a false positive. 46 fixes across 11 reviewed PRs, deployed together.
Designed out a money-transmitter licensing exposure across IL/EU/US before writing code - restaurant as merchant of record, platform fee separate. Read the provider's full API documentation page by page; one buried error code invalidated the refund-retry design.
Led a six-developer team migrating legacy Java applications to Kotlin with modern architecture patterns - improving application efficiency by up to 60% across products used on Android, iOS, and smart-TV platforms.
Designed Python automation and business-growth systems for consulting clients - increasing profile exposure and interactions by up to 800%, built on the same reliability discipline as everything else here.
A restaurant reservation & operations SaaS: ten Python microservices on AWS, event-driven workflows, payments, GitOps delivery, and staff apps from a single Kotlin Multiplatform codebase - designed, built, and operated by one architect.
Led a six-developer team delivering consumer mobile and smart-TV applications for a national telecom - owning delivery lifecycle, technical reviews, mentoring, and architecture modernization across Kotlin, Swift, and Java stacks.
Backend, mobile, web, cloud, and automation systems delivered across industries - selecting the stack for the product, not the trend, and recruiting and mentoring remote teams when delivery demanded it.
Yes. AI can accelerate implementation, but it does not prove authorization boundaries, payment behaviour, rollback safety, or secret handling in production.
Test against the live provider contract, exercise retries and refunds, and prove the intended idempotency and rollback behaviour before release.
Owns architecture and high-risk technical decisions a few days a week, with a written decision log that stays with the team.
The product, the release you are considering, the team size, and the decision or failure mode that concerns you most.
NEXT STEP
A 30-minute call is enough to know whether I can help. If I can't, I'll say so - the same way I declare NO-GO on my own releases.
Prefer async? Email works. I respond within one business day, Israel time (GMT+3).