The MVP was the easy part.Production is engineering.
Practitioners and technology enthusiasts, with years inside strong product companies. We ship fast MVPs, harden what works, and advise boards and CTOs when the stakes are high.
MVP in 21 days · AI-boosted
Thin slice first. Harden only after users prove it.
Experts only
We bring top talent to every engagement.
Results in 5 days
We start with low-hanging fruit, then go deeper. First findings after five days.
Always fixed scope
You know the price up front and decide the next step yourself.
Where we are specialists
Five areas. We turn down work outside them.
A working product in 21 days
We ship a thin slice in three weeks and put it in front of users. If it sticks, we rewrite the solution for scale, so the bigger investment only comes after proof that it works. If it does not, you stop before paying for a full build.
- Scoped MVP in 21 days, not months
- Real users before architecture theatre
- Rewrite for scale only after a clear go
Scalable architecture, plus performance audits
We design systems that survive the next 10x, and we audit what is already slow. Coupling, queues, capacity models on the build side. Flame graphs, query plans and p95 budgets when something is already in production.
- Service boundaries, queues, idempotency, capacity models
- Performance audits: p95 budgets, N+1 hunting, query plans
- Before/after numbers from your real traffic
The boring holes are the ones that get used
Nobody attacks your cryptography. They find the endpoint that forgot ownership checks, the token that never expires, or the public bucket from two years ago.
- Auth and per-object authorization review
- Secrets, key rotation, least-privilege IAM
- CVEs triaged by reachability, not scanner noise
Infrastructure should be boring and reproducible
If standing up an environment is a two-day ritual one person knows by heart, that is a business risk. Everything goes into code, with a restore you have actually run.
- Terraform / IaC, no hand-built snowflake environments
- Canary deploys and autoscaling on real signals
- Backups with a measured RPO and RTO
Advice for boards and technical directors
We sit with founders, boards and CTOs on the decisions that are expensive to undo: build vs buy, rewrite vs harden, hire vs partner, launch vs wait. Plain language, with the engineering trade-offs attached.
- Architecture and delivery reviews for the board
- Build vs buy and make vs partner scorecards
- Hiring plans and capacity forecasts before a launch
We use AI heavily, but we don't let it make the decisions.
It speeds the draft. A person owns architecture, review and the pager.
What you actually get back
Not a slide deck. Findings with evidence, severity and effort. Redacted sample below.
fivarto audit --repo acme/platform --depth fullcodebase
43k LOC · 6 services
surface
118 endpoints
elapsed
5 working days
to clear
9.5 days of work
- sev1api
GET /api/v1/orders issues 1 + 412 queries per request. p95 4.9s under load.
1d - sev1auth
JWTs accepted without an audience check. One tenant's token works for every other.
0.5d - sev1iam
CI deploy role holds wildcard permissions on production.
0.5d - sev2postgres
events(created_at) has no index. Sequential scan over 6.1M rows on every dashboard load.
2h - sev2deps
14 packages with known CVEs. 3 reachable from request handlers.
1d - sev2data
Nightly backups have never been restored. Recovery time unknown.
1d - sev3ci
Test suite exists but no checks are required on main. 6 of the last 40 merges were red.
2h
3 sev1 · 3 sev2 · 1 sev3
You keep the report whether or not you hire us for the fixes.
How an engagement runs
Start with a fixed first pass. Continue only if it earned the next step.
- 01
Day 0 · 45 minutes
A technical call, not a discovery workshop
You show us the repo, the deadline or what is on fire. We say if we are the right team, including when you do not need us yet.
out: go / no-go + which engagement fits
- 02
First engagement · fixed scope
A concrete first pass, matched to the problem
Rapid MVP, architecture or performance audit, security review, infra fix, or advisory for the board. Always fixed scope and price. Early results decide whether we continue.
out: live slice, findings or decision record
- 03
Next · if it earned a continue
We do the rest of the work that matters
Harden the MVP to production-ready, clear SEV1s, rebuild the request path, stand up infra, or stay as architecture partner. Same gates, same pull requests your team reviews.
out: merged PRs, dashboards, rollback, ADRs
- 04
Ongoing · optional
Stay only while it still pays for itself
Design reviews, capacity planning, hiring help, board and CTO advisory. A day or two a week, written down. You can stop after any stage.
out: retainer you can cancel
Things we built and still operate
We run these ourselves. That is why we have opinions about on-call and cost per user.
Pdfaxe
PDF manipulation and extraction. Heavy jobs run in a worker pool so uploads never block the request thread.
Spacelize
Spatial data management. PostGIS indexing and server-side tiling keep large maps interactive.
Pictionary.it
Real-time multiplayer drawing. Authoritative websocket state plus AI content moderation.
TheLifeBand
Senior care wearable: fall detection, vitals, GPS. At-least-once ingestion end to end.
Talleo
Personalised audiobooks for children. Pre-rendered and cached so cost per story stays predictable.
Four partners. No bench.
We orchestrate the engagement end to end: scope, experts, delivery and sign-off.
Oskar Polak
Managing Partner · CTO
Sets the technical direction, owns architecture decisions, and stays accountable for delivery from first commit to production.
Damian Wieliczko
Partner · Head of Analysts
Leads rescue work. Refactors systems while they stay in production.
Marcin Borowiec
Partner · Head of Product
Turns a vague goal into a scope an engineer can build.
Paweł Cegła
Partner · Head of Growth
Partnerships and commercial terms, including equity deals.
expert network
Experts are matched from our network to the specific problem and project. The right people on the right work. That is what makes the quality stick.
Bring us the repository.
Forty-five minutes, screen shared. Tell us what is slow, fragile, or needs to launch. Honest read, including when you should not hire anyone yet.