The problem
Traditional IVR phone trees are slow and frustrating: members navigate rigid menus to do simple things like check a balance, move money, or pay a bill, and anything unusual dumps them into a hold queue for an agent.
Replacing that with conversational AI means the agent has to authenticate the caller securely, understand free-form speech, and actually execute against the institution's real core banking data — not just chat.
The approach
We built conversational voice self-service demos for financial institutions where the caller speaks naturally and the AI performs the banking action through tool calls to the backend, which in turn talks to the core banking system. The voice platform owns telephony, speech recognition, and synthesis; the backend translates the agent's structured tool calls into core-banking API requests and back.
One build targets a retail/credit-union banking platform with biometric (push-to-phone) authentication and a broad catalog of banking tools, backed by stateful stubbed APIs that mirror the real endpoints so they can be swapped for production. Another targets credit unions on a specific core-banking platform, integrating over that core's XML API and adding a cross-channel engine that notices when a member started something in online or mobile banking and offers to finish it by voice.
Both use domain-correct language (members, shares, loans — not customers, accounts) and hand off to a live agent with full conversation context when needed.
What I built
- Conversational banking IVR with simulated biometric push-to-phone authentication and ~15 banking tools (balance, transfers, bill pay, card management, loan payments, ATM/branch locator, alerts, disputes)
- Stateful in-memory banking API modeled on the real vendor interfaces (balances cascade correctly across operations in a call) so stubs can be swapped for production endpoints
- Credit-union voice self-service integrated with the core banking system over its XML API, exposing member, share, and loan data in real time
- Cross-channel engine that detects an action begun in online/mobile banking and proactively offers to complete it by voice, plus CRM interaction logging
- Agent-escalation flow that packages the full conversation for a screen-pop hand-off
- Backend that never touches audio — cleanly translating voice-platform JSON tool calls to core-banking requests
- VAPI
- ElevenLabs Conversational AI
- Twilio
- Python / FastAPI
- Fastify / TypeScript
- PostgreSQL
- Redis
- core-banking XML API
Result
Working conversational voice self-service demos that authenticate callers and complete real banking transactions against core-banking APIs, positioned as a natural-language replacement for legacy menu-tree IVR.
- One demo deployed on production hosting with a full transactional flow (auth, balance, transfer, bill pay) and 22 automated tests
- Stubbed banking APIs are interface-compatible with the real vendor endpoints for a clean swap to production
- Credit-union build integrates over the core's real XML API and adds cross-channel and CRM integration