Work
What we have built, and what it is doing now.
What we have built for customers, the architecture behind our own products, and a public delivery record going back eight years.
01 · Flexiflo
A quality system running on the shop floor.
Flexiflo Corp manufacture hose assemblies out of their facilities in the UAE, Saudi Arabia, Kuwait, Qatar and Oman, ISO 9001 and API Spec Q1 certified. Every assembly they ship carries a test certificate, and every certificate is the end of a chain: a job order, a worker, a tester, an inspector, and a signature from each.
We built the system that runs it now. It carries production orders and workflow templates the quality team can change themselves, and it gives each of the seven roles only the steps that role is allowed to perform. Anyone on the floor or at the test bench can raise a non-conformance report, which a supervisor then reviews and decides on. It produces the final inspection reports and the test certificates — and a customer reaches the certificate for their own hose by scanning the QR code printed on it.
02 · liqwik
A marketplace for unpaid invoices, where the money has to be exactly right.
liqwik is a marketplace where a business waiting to be paid can turn an unpaid invoice into cash now. It lists the invoice; buyers bid for it at a discount, whole or in parts; the seller accepts; the deal becomes an e-signed contract; and the system then follows the original debtor paying it off. Four parties are involved, and each sees a different view of the same deal.
03 · Our own architecture
The parts we built once and reuse.
Each of these was built for one system under a real deadline, then earned its way into the others.
The validated core and the AI layer
Quality software has to be validated, and every change inside the validated part means validating it again. So we made that part as small as we could: document control, approvals, e-signatures and the audit trail, and little else. The AI sits outside it. It can read everything, but the only place it can write is a holding area for proposals. The single way anything crosses from there into the record is a person signing it, and the record keeps both facts — that the AI drafted it, and who approved it.
Consent and a tamper-evident trail
Clinical records need a stronger guarantee than a list of who did what: you also have to show the list itself was not rewritten later. So each entry is sealed against the one before it, and the entries are kept in separate monthly sets per hospital. A check walks the whole chain and reports anything missing or altered, and where — not just a pass or a fail. Consent is a live record the system reads before it acts, rather than a tick box on a form, and if it cannot confirm consent it stops.
Rules an agent drafts, a person promotes
Writing one of these rules was never the slow part. Showing that a new rule does what you meant, against data that already exists, was. So the agent pulls real sample records, writes the rule and the tests for it together, runs them, compares what came out against what should have come out, and tries again. What it hands back is a draft. It then takes the same approval path a hand-written rule takes, and runs boxed off in the same way. The agent writes; it does not decide what goes live.
04 · Before Fractorize was a team
A public record, eight years long.
Before Fractorize was nine people it was one engineer on Upwork: Top Rated Plus, five stars on every project, eight consecutive years. The reviews are public and were written by the clients themselves.
They are about estimates that held, replies that came, and work that was still running years later.
“My team has worked with Manokaran for over 7 years… He provided reasonable estimates and very effectively stuck to those estimates. I foresee that the tools he has built for us are going to be in use for many years to come.”
“Completely capable of tackling very complex issues independently, leading a team, prompt with communication, and thinks deeply about the problems.”