AI for banks: voice agents, KYC checks and private LLMs
We build AI for banks and fintech companies that must keep data inside their perimeter. Collection calls, KYC document checks, anomaly detection and internal assistants run on private LLMs, with a full audit trail.
Typical tasks
- 01Overdue debt collection calls
A voice agent calls debtors on schedule and within the regulator’s rules. It states the amount, agrees a payment date and writes the promise into the bank’s CRM. We built this for IdeaBank under National Bank of Ukraine rules.
- 02KYC document verification
The AI reads passports, ID cards, bank statements and proof of income. Fields are matched against the application, and mismatches go to a human reviewer. The final decision always stays with your staff.
- 03Internal assistant for employees
A private ChatGPT answers questions from internal policies, tariffs and product sheets. Every answer links to its source document. Access follows your existing roles.
- 04Fraud and anomaly detection
The model flags unusual transactions, suspicious applications and repeating patterns. It does not block anything on its own. An analyst gets a signal with an explanation and decides.
- 05Contact center call analysis
Our own LLM runs on dedicated servers and reviews every call. You see script breaches, complaints and risky wording. Recordings never reach third-party models.
- 06Private LLM inside your perimeter
We deploy the model on your servers or on a dedicated server reserved for you. Logs, keys and access stay under your control. The other use cases run on top of it.
Data and access requirements
- Banking regulation: in Ukraine, National Bank of Ukraine rules on data protection and debt collection; in the EU, DORA and the EBA Guidelines on outsourcing arrangements. Your compliance team signs off scripts before launch.
- Bank secrecy and GDPR: NDA and DPA are signed before any access. We work under ISO/IEC 27001:2022.
- Data residency: processing stays in your perimeter or on dedicated servers, with no calls to external model APIs.
- EU AI Act: creditworthiness assessment and credit scoring of individuals are listed as high-risk, so we document data, logic and human oversight from day one.
- Integration with the core banking system, CRM and telephony through API, SIP or file exchange.
What we build for this industry
Case
- // BANK · VOICEIdeaBank
Voice agent for overdue debt collection
TaskReminders about overdue payments took hours of call center operators’ time.
What we didA voice agent runs soft and hard collection under National Bank of Ukraine rules. Each call result is written into the bank’s CRM.
Questions
Can a voice agent follow collection regulations?
Yes. Allowed calling hours, contact frequency, mandatory phrases and banned wording are built into the scenario. Your compliance team approves the script before launch. Every conversation is recorded and available for review.
Where is customer data stored?
On servers inside your perimeter, or on a dedicated server under contract. Recordings and documents are not sent to third-party models. We hold ISO/IEC 27001:2022 certificate No. 4160097.
What does a project cost and how long does it take?
Turnkey projects start at $5,000. Discovery takes 1-2 weeks and the fixed estimate is ready in 2-3 days. The build takes 4-8 weeks with weekly demos. The price depends on integrations and infrastructure requirements.
What if the AI makes a mistake?
AI sometimes gets things wrong, so critical decisions stay with people. The agent hands difficult calls to an operator, and mismatched KYC cases go to manual review. We launch on part of the traffic first and compare with your current process.
Do you work with banks outside Ukraine?
Yes, we work with clients in Ukraine, Poland and the EU. We sign a DPA under GDPR, support on-premise deployment and adapt scripts to local collection and consumer credit rules.
Describe the task. The first call is with the founder.
We reply within one business day. 30 minutes to tell you whether it is worth doing and what it costs.