pixobit
pixobit
services
Services

Six things we do properly

Commerce platforms, ERP and operations software, and the AI layers now built on top of them. Fifteen years spent on the parts that have to be right — stock ledgers, money paths, tenant boundaries — not only the parts that have to look right.

pixobit

Every line is a system we have shipped

Not a menu of things we could learn. Each of the six below is running in production somewhere right now, under a business that would notice within the hour if it stopped.

01 Commerce and marketplace platforms
Storefronts, marketplaces, print-on-demand

A store the business owns outright

Built on OmniCart, the platform we wrote and still maintain. Variable-attribute catalogues, tiered price lists, vendor approval routing and thirteen payment gateways out of the box — multi-currency and multi-language from the first day, not bolted on in year two.

  • Storefront, cart and checkout
  • Vendor onboarding, commissions and payouts
  • Faceted search with live result counts
  • A merchant dashboard non-developers can run
OmniCart CodeIgniter 4 MySQL Stripe
02 ERP and operations systems
Inventory, procurement, production, POS

One schema, two businesses, no spreadsheets

Multi-warehouse stock ledgers with batch traceability and FEFO costing, purchase orders with approval chains and partial receipts, work orders that post inventory only once output is verified, and returnable-asset exchange modelled as a real order state rather than a note in a column.

  • Stock ledger, batch cost allocation, stock takes
  • Purchase orders, approvals, supplier costing
  • Work orders and production runs
  • POS that keeps selling when the line drops
Fluxion ERP PHP 8.4 MySQL Redis
03 AI on your own data
Retrieval, agents, guardrails

Answers grounded in your records, not the internet

Assistants that read your own corpus, query tools scoped so hard that a generated query crossing a customer boundary is rejected before it runs, and task-specific agents — analyst, translator, SEO, content — sharing one set of guardrails instead of a prompt per feature. We gave the keynote on this at AI & Tech Manila.

  • Retrieval architecture over internal documents
  • Tenant-scoped SQL and schema tools
  • Multi-agent task pipelines
  • Evaluation, logging and cost control
RAG OpenAI API Tool calling Elasticsearch
04 Workforce and field operations
Attendance, shifts, payroll, tracking

The people and vehicles, on the same ledger

Radius-validated punches against a geofence, shift duties and approval workflow, statutory payroll calculation, and live map tracking of drivers and deliveries — all reading one geo-ping ledger instead of four apps that each hold a different version of the day.

  • Geofenced attendance and duty scheduling
  • Approval workflow and exception handling
  • Payroll runs with statutory tax rules
  • Driver and delivery tracking on a live map
Leaflet MySQL Payroll rules
05 Rescue, performance and architecture
Someone else’s codebase, still in production

Fixed against live traffic, without a maintenance window

We reshape the access patterns behind the worst queries instead of papering over them with indexes, move write-heavy work such as audit capture off the request path, and replace polling and manual handoffs with event-driven workflow. Seven years of doing exactly this on one platform, with no downtime.

  • Query and access-pattern rewrites
  • Data modelling and schema repair
  • Event-driven workflow automation
  • Handover documentation and a regression suite
Profiling PHPUnit Zero downtime
06 Hosting and maintenance
Monthly retainer, servers we manage

Someone who already knows the codebase

Backups, security updates, performance monitoring and bug fixes on a fixed monthly rate — answered by the people who wrote the thing, not by a queue. The source code stays yours either way, so the retainer is a choice rather than a lock-in.

  • Database backups and restore drills
  • Security updates and dependency upgrades
  • Uptime and performance monitoring
  • Small changes, without a new quote each time
Linux Docker CI/CD
pixobit
Engage

Three ways this usually starts

Most enquiries land in one of these. Tell us which one sounds like you and the first conversation gets a lot shorter.

01

Build it from zero

You have a business and a problem, and nothing to start from. We do the discovery, the data model, the design and the build — staged so something useful is live long before the whole thing is finished.

02

Take it over

There is a system in production and nobody left who knows it. We read it, map it, stabilise the worst of it and then run it. Two of the platforms we have inherited this way, we still maintain years later.

03

Add the layer you are missing

The platform works, it just does not do the new thing yet. An AI assistant over your own data, a marketplace side, an integration, an API for a mobile app — built onto what you already run.

15+ Years in production Shipping software other people depend on, across three countries and several stacks that have since gone out of fashion.
20+ Systems delivered ERP, commerce, warehouse, booking, ticketing and automation — architecture and maintenance owned, not handed off at launch.
147 Migrations on one live schema A single ERP: 49 domain entities, 67 models, 147 versioned migrations, applied to a business that never stopped trading.
7 Years on one platform Our longest single engagement. Long enough to live with every decision we made in the first year.
pixobit

The half of the job you never see

Everything that decides whether a system survives its second year is invisible from the outside. It is also the part we are actually hired for.

Ledger invariants

Stock and money move through one path with rules that cannot be quietly bypassed, so the books still balance a year and four features later.

Concurrency and oversell

Two customers, one last unit, the same millisecond. That answer has to be decided before launch, not after the first refund.

Tenant boundaries

Scope resolved from the record that owns the work, never from whoever happened to trigger it — so cron jobs, queued work and offline replay all post to the right customer.

Event-driven workflow

Rules react to what happened instead of polling for it, and a slow side effect never blocks the transaction that set it off.

Tests on the money paths

Cost allocation, stock availability, sourcing policy, payroll runs and order ledgers are covered by a suite that fails loudly the moment someone changes the rules.

Migrations you can run forward

Schema change is versioned and reversible, which is what turns a deploy into a routine afternoon instead of a night operation.

Books that balance Nothing oversold Your data stays yours Deploys without downtime Money paths covered by tests Schema changes you can undo Fixes that do not break something else Still standing in year two
pixobit

The questions everyone asks

Answered here so the first call can be about your business instead of ours.

How do you price the work?

Fixed price per stage where the scope is genuinely clear, and a monthly rate where it is not. Part of the first conversation is telling you honestly which of the two your project is — quoting a fixed price for discovery work nobody has done yet is how projects end badly.

Who owns the code?

You do. Source included, no cut of your transactions, no licence that expires. If you ever leave, you leave with a running system and the means to keep running it.

Can you work with our existing developers?

Yes, and most of fifteen years has been spent inside code somebody else wrote. We can lead the architecture, take one hard subsystem, or sit in your review process as another pair of hands — whichever the team actually needs.

How long does it take?

A storefront on OmniCart is measured in weeks. An ERP that has to be right about money and stock is measured in months, staged so the first useful piece is live early and everything after it lands on something already proven.

Do you design as well as build?

Yes. Look and structure are worked out with you before anything is built, and you follow the build on a staging environment from day one. If you already have a design, we skip straight to building it.

What happens after launch?

Four months of support are included — fixes, questions and small changes. After that a monthly retainer is available, and it stays optional. We would rather you keep us because the work is good.

Where are you, and in what language?

Târgu Mureș, Romania, working remotely with clients across Europe, the United States and the Philippines. English, Hungarian and Romanian.