Fintech, Nepal
A digital payments company in Nepal scores fraud in real time
A digital wallet provider in Nepal streams every transaction through DataNature StreamHouse and scores it for fraud as it happens. Models train on governed lakehouse history, and the whole platform runs in Nepal to meet payment-data residency expectations.
- Customer
- A digital wallet and payments company in Nepal
- Country
- Nepal
- Hosted on
- In Nepal
- Products
- DataNature, AI360 Studio
The challenge
Transaction volumes grew fast. Fraud rules ran in batch overnight, so losses were found the next day. Partner banks asked for clear answers on where data lived.
The solution
Connect captured changes from the payments database into StreamHouse. Fraud models scored each event in real time. Atlas gave partner banks lineage and access evidence.
The results
- Fraud scored per transaction, not per night
- Lineage and audit evidence for partner banks
- Payment data kept in Nepal
What was the company trying to fix?
Overnight fraud rules were too slow for instant payments. Partner banks wanted assurance on data residency and access.
What did it deploy?
DataNature StreamHouse for real-time events, the Lakehouse for model training, AI360 Studio for scoring, and Atlas for audit.
Why it worked
The same governed tables serve real-time scoring and model training, so there is one version of the truth for fraud, finance and partners.
How does StreamHouse power real-time fraud scoring?
This is the canonical StreamHouse hot path:
| Step | What happens |
|---|---|
| Capture | Change data capture from the payments database streams every top-up, transfer and merchant payment into StreamHouse |
| Contracts | The schema registry fixes the event format that scoring depends on |
| Features | In-flight processing keeps short-window features per wallet and device: transfers per minute, new beneficiaries, device sharing across wallets |
| Score | A fraud model from the DataNature AI registry scores each event in real time, alongside rules for known mule patterns |
| Act | Suspicious transfers are held or stepped up; linked devices can be frozen across wallets |
| Evidence | Events, scores and decisions land in Iceberg; Atlas lineage and access logs answer partner banks' questions on where data lives and who saw it |
Fraud that batch rules found the next morning is now flagged while the transfer is in flight. Slower features, such as account age and historical behaviour, come from nightly batch models in Gold.
This story is anonymized at the customer’s request.
Platform components in this story
Learn more about each DataNature product and module mentioned here.
Products
See it on your own data.
Book a demoDemos and quotes are handled by Dlytica, the company behind DataNature.