Retail
Banker.
A suite of narrow models for retail banking applications. Each model is trained on the bank's own data, removes a specific friction from everyday banking, and runs entirely inside the bank's own infrastructure.
Six models. Six frictions removed.
Merchant Intelligence
Classifies and enriches transaction data so a statement reads as a list of recognisable places rather than a list of processor codes.
Pay with Photo
Extracts payment details from an image of a bill or invoice, reducing the keystrokes between a bill arriving and a payment being ready to confirm.
Subscriptions
Identifies and surfaces recurring payment commitments so customers can see all of them in one place, including those that have drifted out of mind.
Pay from Clipboard
Detects payment details that have already been copied to the clipboard and pre-populates the payment form, removing the need to type them again.
Salary Awareness
Detects salary patterns and surfaces products and timing recommendations that fit how a customer is actually paid.
Workflow Automation
Replaces repetitive internal banking processes with consistent, auditable automated workflows trained on the bank's own procedures.
How Retail Banker is built
Each model in the suite is narrow by design — trained on a single task using the bank's own transaction data, document history, and internal records. No customer data passes through a third-party API at any point during training or inference, and the resulting weights are owned by the bank.
None of the models move money or make a credit decision. They prepare, classify and enrich; the bank's existing controls and the customer's own confirmation do the rest. This boundary is what makes deployment defensible under FCA regulation. Our wider position on oversight and evaluation is set out in our responsible AI statement.
For full details on data handling, weight ownership, and sub-processor arrangements, see our data and model ownership statement.
The base adapter is published on HuggingFace: NavitraTechnologies01/bn-01-retail-banker (LoRA adapter on Qwen3-14B).
Deployed inside your infrastructure
The models run on the bank's own servers or private cloud. Customer data never leaves the bank's environment during training or inference.
Weights owned by the bank
The trained model weights belong to the bank, not to Navitra. The bank can audit, retrain, or decommission them without any dependency on us.
Modular — adopt one or all
Each model in the suite is independent. Banks can deploy a single component alongside existing systems or adopt the full suite in phases.
FCA-aware design
No model in the suite makes a regulated decision. All outputs are presented to a human for confirmation before any action is taken.
Interested in Retail Banker
for your bank?
Request the technical brief — architecture, training methodology, and evaluation results shared under NDA.
Request the technical brief