Software Architecture Review
Teams call me before a scaling push, before committing to a rewrite, before a raise, or when shipping has quietly slowed and nobody can name the reason. I read the system you actually have, tell you what breaks first, and hand you the fixes in the order they pay off.
A software architecture review — some teams call it a system design review — is a structured assessment of an existing system: its boundaries, data flows, failure modes, and cost profile, measured against where the business is actually going. Oleg Sotnikov runs these reviews as a fractional CTO with 25+ years in IT, 7 patents, and 1,000+ projects behind him, including the clinic ERP architecture he shaped for QueueStone and stayed with through delivery. You get a written report: what breaks first, what is over- or under-engineered for your stage, and a roadmap sequenced so each fix makes the next one cheaper.
What the Review Covers
Six angles on the same system. Which ones matter most depends on your stage, and we settle that on the first call.
Scalability and bottlenecks
Where the system stops holding at ten times today's load: the hot table, the synchronous call sitting in the request path, the queue with a single consumer. I follow the requests that matter through every hop they make, not the diagram of how they were supposed to travel.
Stage fit
Kubernetes for 200 users costs more attention than it will ever save. One shared database at Series A costs you an outage. I mark both directions: what you built years too early, and what you can no longer postpone.
Security and data flows
Who can reach what, where secrets live, which service implicitly trusts which, and where personal or regulated data crosses a boundary it was never meant to cross.
AI-readiness
Coding agents work well in systems with clear module boundaries, honest tests, and short feedback loops, and flail without them. I assess how much of the codebase an agent can safely touch today, review the parts that were AI-generated, and name the changes that widen that surface.
Cost profile of the design
Architecture decides the bill long before the cloud console does: chatty cross-zone traffic, per-request fan-out, storage that is never tiered, a model call fired on every keystroke. I price the design, not only the invoice.
Delivery impact
Test suites nobody waits for, a build that only runs in CI, one staging environment three teams queue for. Complaints about velocity usually trace back to structure rather than effort, and structure is fixable.
How It Works
Context and access
A call about where the business is heading over the next twelve months, then read-only access to the repos, infrastructure configs, and whatever diagrams exist. NDA on request. Nothing changes in production.
The review
I read the code and the infrastructure, trace the critical paths end to end, and talk to the engineers who carry the system day to day. Scope and timeline are agreed before I start, and they follow the size of the system.
Report and walkthrough
A written report with findings, risks ranked by what they cost you, and a sequenced roadmap. Then a live walkthrough with your team, so the reasoning reaches the people who will implement it.
Why Me
- 25+ years in IT, 7 patents, and 1,000+ projects — architecture decisions I had to live with afterwards, not just recommend
- Shaped the architecture of a mental-health clinic ERP for QueueStone and stayed involved through delivery (quote below)
- AppMaster's architecture serves users in 190+ countries at 99.99% uptime on about $650/month of AWS
“Oleg Sotnikov was a great support to our team as Fractional CTO. He helped us shape the architecture of an ERP system for a mental health clinic and stayed involved throughout the development process, helping us make better and more efficient decisions. We truly appreciate his contribution.”
Frequently Asked Questions
What is a software architecture review?
A software architecture review is a structured assessment of an existing system: how it is split into parts, how data moves between them, where it gives way under load, and what the design costs to run. It is anchored to one company's trajectory rather than to generic best practice, because the right architecture for 500 users and the right one for 500,000 are different systems. The deliverable is a written report with findings, risks, and a fix order.
When do we need an architecture review?
Four moments justify it: before a scaling push, before committing to a rewrite, before a funding round or acquisition where technical questions get asked in detail, and when delivery has slowed and nobody can explain why. A review pays for itself most when it happens before the decision rather than after the migration. If you are already mid-rewrite, it can still tell you whether finishing is the cheaper option.
Do you review AI-generated codebases?
Yes, and that is a growing share of the work. Code produced by agents or by vibe-coding tools tends to be locally reasonable and globally inconsistent: duplicated logic, module boundaries that drifted, tests that assert nothing, dependencies nobody chose on purpose. The review separates what is safe to keep from what needs rewriting before it carries production load, and names the structure to impose so AI tools keep helping instead of compounding the mess.
What does the architecture review report contain?
Findings ordered by consequence: what breaks first, what is over-engineered for your stage, what is missing, and where the design drives your cloud and LLM bills. Each one names the affected component and the work it takes to fix. After that comes a sequenced roadmap — now, next quarter, leave alone — and a walkthrough call with your engineers so the reasoning survives the handoff.
How much does an architecture review cost?
It is scoped per engagement, because a forty-service platform and a single Rails app are different amounts of reading. The pricing page carries the fixed numbers: the $5,000 Team & AI Audit over five business days, and fractional CTO engagements at $5,000–10,000 per month. When the question is broader than architecture and covers team structure, AI adoption, and cloud spend together, the $5,000 audit is the better place to start.
Find Out What Breaks First
A written review of the system you actually have, with the fixes in the order they pay off.
Thirty minutes on a call is enough to scope it. NDA on request.
Related reading
Architecture, scaling decisions, and what AI tooling changes about both.
