Loading…
The studio
A small engineering studio, built for long-lived systems
Mazarix designs and builds the software companies depend on daily — and keeps it simple enough that one team can hold the whole thing in its head.
Most business software fails in one of two ways. It is bought off the shelf and covers most of what you need, leaving the rest to spreadsheets and memory. Or it is built quickly by whoever was available, works for a year, and then nobody can safely change it.
We build for the second decade, not the first demo. That means a data model designed before screens, access rules enforced where they cannot be bypassed, types that catch a mistake at build time instead of on a Friday night, and a deployment you can reverse in a minute.
It also means saying when software is the wrong answer. Plenty of problems that arrive as 'we need a system' turn out to be a process nobody has written down. We would rather tell you that in week one than bill you for six months of building around it.
The founder
Shahram Mazar
Founder, Mazarix
Shahram Mazar holds an MBA and a degree in Computer Science, and has spent more than 15 years writing software. At Mazarix we work on AI, automation, web applications and software built for one business, alongside AI and automation teams in several countries. If your company, workshop or factory needs an application of its own, talk to us and we will say exactly how we can help.
Talk about your workAI · automation · web applications · software built to fit
How we work, concretely
The data model comes first
Screens are easy to change; a wrong data model is not. We spend the first stage on how information actually relates before drawing anything.
Security is structural
Permissions live in the database, not only in the application in front of it — so a mistake in one screen cannot expose another customer's records.
Small, frequent releases
Two-week increments, each deployed and usable. You never wait months to find out whether we understood you.
Written down, not remembered
Decisions, schemas and trade-offs are documented as we go. The system stays understandable after the people who built it move on.
Boring technology on purpose
We choose tools with long support horizons and large communities. Novelty is a cost you pay later, usually at the worst moment.
Clear ownership
The software, the data and the documentation are yours. Where it is hosted and who has access is agreed in writing before the project starts.
How we work
What working together looks like
Short, defined engagements with a clear end — not an open retainer that quietly becomes permanent.
01
Understand
We map how the work moves today — who touches what, where it stalls, which systems disagree. You get a written picture of the process before anyone proposes software.
Process map · scope · estimate
02
Design
We decide the data model, the architecture and the screens together. Where a simpler answer exists — an automation instead of a platform — that is what we recommend, even when it is the smaller project.
Technical spec · data model
03
Build
Staged delivery in two-week increments, each one deployed and usable. The task causing the most pain ships first, so the work gets easier before the project ends.
Working increments · staging
04
Hand over
The system is deployed on the agreed infrastructure. You get administrative access, documentation, and a training session for your team.
Production · docs · training
Discovery
One to two weeks. We map the process, agree scope, and give you a written estimate. If we do not think the project is worth doing, this is where we say so — and you owe us nothing further.
Build
Two-week increments with something usable at the end of each. You see progress continuously and can change direction before it becomes expensive.
Hand-over
Deployed on the agreed infrastructure, documented, with a training session for your team.
After
Support, monitoring and new features are separate agreements you opt into. Most clients take a small monthly arrangement; some take nothing and call us when they need something.
We are probably a good fit if
- You are running operations on spreadsheets that no longer agree with each other
- You bought software that covers most of the job and cannot be changed
- Someone on your team is manually moving data between two systems every day
- You have an idea for a product and need it built properly the first time
- You want AI applied to something specific, not AI applied to everything
And probably not if
- You need someone to maintain an existing site nobody can read
- The main requirement is the lowest possible price
- The process is still being argued about internally
- You want a team to embed full-time and take direction day to day
Straight answers
The questions people actually ask
How long does a project take?
A focused automation is two to four weeks. A full operational system is usually three to six months, delivered in two-week increments you can use as they land — you are not waiting until the end to see it.
What does it cost?
It depends on scope, and we will not pretend otherwise. After the first stage you get a written scope with a fixed estimate, before any build work starts. If the work is not worth doing, we say so then.
Do I own what you build?
You own the running software, the data and the infrastructure it sits on, with full administrative access. Source code transfer is a separate arrangement — ask and we will price it.
What happens to my data with AI?
Before any AI work begins we put in writing which data leaves your organisation and where it goes. Your data is never used to train a model and never appears in another client's project. Where sensitivity demands it, running the model on internal infrastructure is assessed and implemented if feasible.
Can you work with the systems we already have?
Usually. Anything with an API, a database or even a scheduled file export can be connected. Where a system is genuinely closed, we will tell you before you spend anything.
What if we need changes later?
You get an admin panel for the things that change often. Beyond that, new features are a separate piece of work with their own estimate — never an open-ended retainer you forgot you were paying.
Live figures from this very website
Transparency
Live figures from this very website
The system showing you this page was built the same way we build for you. The figures below were read from its own code and database just now; they are not a mock-up.
Last commit on main: 159c107
The last run, before releaseEvery one must be green; if one goes red, nothing ships.
Typecheck typecheck
no errors
TypeScript, from the database to the screen
Logic tests logic suite
16,143
passed · 0 failed
Database tests SQL · RLS
626
assertions green · 67 files · on a real Postgres · each customer's data isolated · 51781e1
E2E tests end-to-end
1,681
passed · 0 real failures · on 4 devices · 10e7e01
Security & penetration security · pentest
91
Attacks written down as tests: SSRF, SQL injection, rate caps, a security probe on every device
Performance budget query budget · drift
held
No page's query count has grown · the schema matches the code
13 device profiles; the 4 highlighted ran last:iPhone SEiPhone 14Pixel 7Laptop 1280iPhone 15 Pro MaxGalaxy S9+iPad MiniiPad Mini landscapeiPad Pro 11Desktop 1440Safari 1512Window 1024
What is under this pagecounts, from the repository
The repository itself (GitHub Repo)from git, counted at 399b43e5
816
commitssince 27 August 2026
69
commits in the last seven days
224,874
lines of TypeScript
38,965
lines of SQL in migrations
1
person, every commit signedShahram · Mazarix
96
documents the assistant answers from, with a citation on every reply
81.7 / 10
score of the last graded run, across 63 test cases
177 migrations, in the order they were appliedone tick per schema change · the six tall ones are turning points
0001Foundation0004Booking0009Assistant0012Email queue0032Email, once0054Draft schedule0095Proposals and contracts
- 0001FoundationExtensions and helpers; the ground everything else was built on.
- 0004BookingThe bookings table, and the rule that no two overlap — in the database, not the screen.
- 0009AssistantSemantic search over the site's documents; every answer cites its source.
- 0012Email queueEmails are queued and sent from the queue, not in the moment.
- 0032Email, onceThe fix for a confirmation that went out three times; every send carries a unique key since.
- 0054Draft scheduleThe assistant's draft replies, scheduled and reviewable.
- 0095Proposals and contractsA proposal the client answers from a link, and the contract they sign electronically.
Remote, on Atlanta time
Mazarix works remotely with clients across the United States and internationally. Consultations run over Google Meet. Eastern time, in English or Persian.
There is no walk-in office, and everything starts on this site: the contact form, or the free membership. We put the reply in your account and send a copy to your email.