Private AI, one model per mission.

Extracts and checks identity, company and ownership documents.

Summarises financials, cash flows and risk signals into a draft credit memo.

Tuned on central bank, AML and sanctions rules for each market.

Compares LCs, invoices and bills of lading to find discrepancies.

Answers within approved product terms and fair-treatment rules.

Reads proposals, inspection reports and history to score risk.

Compares clauses, exclusions and endorsements.

Classifies claims by severity, coverage and fraud likelihood.

Flags inconsistent narratives, duplicate claims and unusual patterns.

Prepares treaty and facultative submissions and tracks cessions.

Detects anomalies and fraud patterns in near real time.

Checks sanctions, PEP and adverse-media hits and explains each hit.

Assembles chargeback evidence against card scheme rules.

Assesses merchant onboarding and ongoing behaviour.

Matches settlements, fees and FX across rails and flags breaks.

Writes product descriptions, attributes and translations from raw data.

Forecasts demand from orders, seasonality and promotions.

Suggests prices and promotions within margin rules.

Predicts churn and recommends save offers.

Resolves delays, returns and refunds with the right next action.

Reads drawings, BOMs and technical specs.

Links defects to lines, batches, materials and suppliers.

Balances orders, capacity, materials and changeovers, and replans when something slips.

Scores suppliers on quality, delivery and risk.

Spots early failure signs in asset and breakdown history.

Turns live tracking events into status and ETA updates.

Flags likely delays early and proposes re-routes.

Drafts and checks declarations, HS codes and trade paperwork.

Recommends carriers by cost, speed and reliability.

Balances slotting, picking and dispatch against capacity.

Spots defects, anomalies and missing parts at line speed.

Recognises intrusions, falls, fires and unsafe acts in live video.

Understands what is happening in a zone, such as vehicles, objects, occupancy and blocked exits, without identifying anyone.

Faces and identities are masked by default.

Runs on site, close to the cameras, for low latency and local data. [CONFIRM: on-site running implies self-hosting; self-hosted models are still rolling out. OK to state, or word as rolling out?]

Turns each detection into a case with clips, context and the next step.

Natural chat and voice in English, Urdu and more, with Arabic on the roadmap. [CONFIRM: draft said English, Urdu, Arabic; platform list has Arabic on the roadmap]

Keeps context across channels and sessions, within consent rules.

Detects what the customer needs and when to escalate.

Passes a summarised case to a person with full context.

Answers only from approved knowledge and logs every action to the case.

Knows where a case sits in its value stream and what the next stage needs.

Reads StageContracts and only proposes compliant actions.

Respects Role x Action x Scope and never shows data outside the user's scope.

Spots bottlenecks, rework and SLA risk across cases.

Builds evidence packs for human sign-off, never makes the decision itself.