ميزان — the balance.
Every compliance tool claims to find more risk. We built the one that can show its work — because when a regulator asks why, "the model said so" is the only wrong answer.
It started with six spellings of one name
Our founders were building remittance rails for the Gulf when their compliance vendor's screening flagged nothing — because it matched "Mohammed" against "Muhammad" and found them different. The same vendor then drowned the queue in false hits between "Falcon Trading LLC" and "Falcon Trading FZE".
That gap — between what screening systems did and what the region's names and money actually look like — became Mizan AI. Not a dashboard bolted onto a fuzzy-matcher, but an investigation workspace where every weight is visible, every score is reproducible, and the agent writes its own defence before you need one.
— R. Sharma & L. Al-Haddad, co-founders, 2024
What we refuse to compromise
Nothing is a black box
Every score is a sum of itemised factors. Every factor has a documented delta. If we can't explain a decision, we don't ship it.
Agents assist, people decide
Our agent works through the night so analysts don't have to — but it never files an STR, freezes funds, or closes a case alone. Attribution is mandatory.
Reproduce, always
No stochastic models, no silent retrains. The same inputs produce the same score on any machine, in any year — and the engine version is in the trace.
Built where the names are
Transliteration, corporate suffixing, Gulf typologies — these aren't edge cases to us, they're the product. We built from the data, not from a Western template.
We answer for our work
No outsourced support, no ticket black holes. The people who write the engine read your escalations.
Privacy by architecture
Region-pinned clouds, on-prem options, one-click export. We hold the least data necessary and keep it exactly where you put it.
The people holding the scales
Ex-payments infrastructure lead. Spent three years inside the false-positive problem before deciding to fix it.
Built name-matching research at a Gulf central bank. Wrote the first version of the transliteration engine on a weekend.
Former MLRO at a regional exchange. Turns regulator questions into product requirements — verbatim.
Obsessive about determinism. If a score can't be reproduced byte-for-byte, it doesn't merge.
Keeps us honest: runs every release against real exam questions before customers ever see it.
Believes a case file should read like a ledger — calm, complete, impossible to misread.
Median time to first production screen on her watch: 48 hours. She treats that as a challenge, not a stat.
Keeps the matching stack sharp against new transliteration corpora and list formats.
The ledger so far
Two engineers, one hypothesis
Rohan and Layla build the first transliteration-aware matcher and score it against a year of Gulf remittance data.
The agent starts reasoning
First reasoning trace ships — the moment a machine wrote down why, in plain language. The demo that becomes the product.
First production customers
Two exchanges and a trade-finance house screen live. Combined false positives down 38% in the first month.
goAML export & audit log
Regulator-shaped reporting ships; every case now carries an immutable chain of custody.
SOC 2 Type II · GCC region pins
Audit complete; customers can pin data residency to the GCC, EU or US. On-prem preview released.
The MLRO Console ships
Escalation queue, dual-control approvals, committee review flows, workflow rules and supervisory reporting go live with the Enterprise tier. Explore it →
Come weigh with us.
We're hiring engineers who care about determinism and compliance folks who've seen the other side of an exam. Or just book a demo — the coffee is on us either way.