AI fraud intelligence for regulated finance
TasawurAI is building an explainable AI platform that helps financial institutions detect suspicious behaviour, understand risk, and investigate fraud across increasingly complex digital transaction environments.
Designed around explainability, human oversight and enterprise deployment.
Decision stays with the bank.
Our work sits where financial technology, machine learning and regulatory expectation meet. The platform under development is a decision-support layer: it is designed to help fraud, risk and compliance teams see what is happening across digital transactions and to give them the evidence to act — not to make the decision on their behalf.
We are early. Everything described on this site is what we are building and how we intend it to work.
Digital banking and instant payments have removed the interval that fraud controls were designed around. Detection now has to happen while the transaction is still in motion, and investigation has to happen against a queue that keeps filling.
Rule-based monitoring remains necessary — it encodes what an institution knows and it is auditable by construction. But a rule describes a pattern that has already been observed, and it cannot express degree, context or relationship. Fraud adapts around it, and every threshold tightened to catch more activity also produces more alerts that turn out to be nothing.
The result is familiar to every fraud team: a queue that grows faster than the capacity to work it, and an investigator reconstructing context by hand from systems that were never designed to be read together.
Fraud rarely exists as a single signal.
It exists as a relationship between signals — a device that has never been seen, a beneficiary added minutes earlier, a change in pace, a connection between accounts that no individual transaction record contains. TasawurAI is being built around that premise.
Rules, machine learning, anomaly detection and graph analytics each read the same transaction differently. The engine consolidates those readings into a single position — with the reasoning preserved rather than discarded.
Four independent evaluations of the same transaction, resolving into one position.
How these are sequenced, combined and weighted is our own work, and is not published. What we do publish is the output — one risk position, the evidence behind it, and a person who decides.
Fraud typologies the platform is being designed to address. Each is approached through a combination of detection methods rather than a single technique.
The account is familiar. The person using it is not.
A credential compromise usually shows up as a cluster of small departures before it shows up as a loss — an unrecognised device, a session that behaves differently, a payee that has never been used.
Detection approaches described here are in development. They are not a statement of deployed performance.
Individual transactions can look normal in isolation. Graph analytics helps reveal relationships, clusters and transaction pathways that may otherwise remain hidden.
Fifteen entities, each transacting within its own limits. Reviewed one record at a time, nothing here asks for attention.
Accounts, devices, beneficiaries and identifiers are resolved into entities that can be reasoned about, rather than repeated strings across separate records.
Relationship analysis is designed to surface coordination — several accounts reaching one beneficiary, a device reused across them, activity that only looks ordinary one record at a time.
Graph findings are intended to arrive as evidence attached to a case, so an investigator sees the network around an alert instead of assembling it manually.
A score on its own cannot be reviewed, challenged or defended. Every risk position the platform produces is designed to arrive with the signals that created it and the weight each one carried.
Manual review
A recommendation, not an action. No customer-facing decision is taken by the platform.
Behaviour anomaly
Session and timing outside account profile
Previously unseen beneficiary
Payee added 14 minutes before transfer
Transaction velocity
Fourth outbound transfer within the hour
Related entity signal
Beneficiary shares a device with two accounts
Rule trigger
Amount within 2% of an internal threshold
The platform scores, explains and recommends. Investigation, judgement and the final decision remain with authorised personnel inside the institution — and the record of that decision is part of the system, not an afterthought.
Model
Produces a score
Evidence
Retains what it used
Explanation
States it in reviewable terms
Analyst
Reads, investigates, judges
Decision
Recorded with its reason
TasawurAI is built to work alongside existing core banking and payment infrastructure, and to support different deployment patterns depending on what the institution's environment and obligations require.
Where each part runs is a deployment decision taken with the institution. This is not a claim that customer data leaves, or must remain within, any particular environment.
How these are sequenced, combined and weighted is our own work, and is not published. What we do publish is the output — one risk position, the evidence behind it, and a person who decides.
Institutions evaluating the platform under NDA receive the detail their risk, audit and technology functions need to make a decision. That material is not on a public website.
Detection runs inside the institution's own environment, with customer data remaining under its existing controls.
Dedicated infrastructure provisioned for a single institution, isolated from any shared tenancy.
A controlled interface between bank systems and the detection pipeline, scoped to the attributes detection requires.
Model development and validation supported centrally, with releases moving into the bank environment under institutional approval.
A model that cannot be explained, monitored, versioned and audited does not belong in a regulated institution. These principles are constraints on how the platform is being built.
Every risk score is designed to arrive with the signals that produced it and the weight each one carried, in language a reviewer can act on.
The platform is a decision-support layer. Investigation and the final decision remain with authorised bank personnel.
Alerts, evidence, analyst actions and model versions are intended to be recorded as an immutable trail that can be reconstructed after the fact.
Access to cases, customer data and model controls is scoped to role, so investigation and model administration stay separated.
Model performance is intended to be tracked continuously against live outcomes rather than assessed once at deployment.
Shifts in input distributions and in fraud behaviour are monitored, because a model's accuracy is a function of the world it was trained on.
Every model, feature set and rule configuration is versioned, so any historical decision can be tied to the exact logic that produced it.
Model changes are designed to move through validation and institutional approval before they affect production decisions.
The platform is designed to process the attributes detection requires, rather than to accumulate customer data by default.
Data protection is treated as an architectural constraint — deployment topology, access boundaries and retention are decided with the institution.
TasawurAI is pursuing controlled validation through the Palestine Monetary Authority Regulatory Sandbox framework. Our proposed testing approach is structured around controlled validation, human oversight and regulatory alignment.
This describes our intended approach to testing. It is not a claim of approval, endorsement, certification or partnership by any regulator or supervisory authority.
Interface concepts for the fraud, risk and compliance teams who would use the platform daily — designed for density and traceability rather than for dashboards that look impressive from across a room.
34
+6 last hour
12
4 analysts
5
2 to compliance
21
9 confirmed
| Reference | Time | Amount | Channel | Score | Band | Status |
|---|---|---|---|---|---|---|
| TX-82719 | 14:22:07 | 12,400.00 | Mobile | 82 | High | Open |
| TX-82704 | 14:19:51 | 3,150.00 | Web | 74 | High | In review |
| TX-82688 | 14:16:33 | 890.00 | Mobile | 61 | Elevated | In review |
| TX-82671 | 14:11:02 | 22,000.00 | Branch | 58 | Elevated | Open |
| TX-82650 | 14:04:48 | 410.00 | Mobile | 44 | Elevated | Assigned |
| TX-82612 | 13:58:19 | 1,275.00 | Web | 27 | Low | Closed |
Interface concepts in development. All references, amounts and scores shown are sample data.
TasawurAI works at the intersection of financial technology, artificial intelligence, fraud intelligence, risk and regulatory technology — with a first focus on the Palestinian banking sector and an architecture intended for regulated institutions more broadly.
Building more secure, transparent and resilient digital financial ecosystems.
Providing regulated institutions with reliable, explainable and regulator-conscious fraud intelligence.
Co-founder
Works across AI and machine learning systems, document intelligence, software engineering and banking technology, with a focus on how models are built into production software rather than left in notebooks.
BSc Computer Science & Artificial Intelligence, IE University, Madrid
Co-founder
Works on production financial fraud detection, graph-based fraud modelling, transaction analytics and machine learning deployment within international financial organisations.
BSc Computer Engineering · MSc Artificial Intelligence Systems
More about the company on the company page.
Financial systems are becoming faster.
Fraud is becoming more connected.
Risk infrastructure must become more intelligent.
We’re building toward that future.
TasawurAI
For financial institutions, technology partners and regulatory stakeholders interested in TasawurAI, get in touch with our team.