AI-first online banking platform

The product offers credit and banking services to customers who arrive through partner platforms such as Uber and through CRM and WhatsApp conversations. It was designed AI-first: conversational AI does the first contact and the onboarding work, while financial decisions stay in a core platform with explicit rules.

An external conversational layer on the CRM and WhatsApp handles first contact. It hands the customer to an internal AI that completes onboarding, and the internal AI calls the core platform, which owns credit analysis and simulation, anti-fraud, formalization and digital signature, and the integrations around them.

The core is a modular monolith with hexagonal architecture rather than microservices: each domain module keeps its business rules free of framework code, and use cases depend on ports, not on the database or the message broker. That keeps one deployable unit while the boundaries stay enforceable.

AI is an accelerator, not the decider. Guardrails redact sensitive data before any model or conversation history sees it, can block a step, and log only safe technical metadata.

CRM events reach the core through event-driven ingestion that can be replayed, and observability and feature flags were part of the platform from the start.

I work on the core platform and built the partner landing page: mobile-first and mostly static, with SEO and Core Web Vitals treated as product scope, plus an alternative build for the CRM's marketing pages. I integrate the core with the internal and external AI layers, which colleagues build.

The company, its clients and data, internal names, metrics, code and screenshots stay private under confidentiality terms. The diagram above is a simplified illustration, not the production topology.