Building the Future of Banking

The Challenge

Alpian aims to provide private banking-quality service at a fraction of the traditional cost. As part of this mission, they wanted to create a customer-facing conversational AI chat agent for their digital banking platform. This would allow users to ask questions about their Alpian account data using conversational language, while freeing up human advisors to focus on complex wealth management conversations.

Additionally, Alpian wanted to offer new advanced features, such as geolocation intelligence to track spending by location and financial insights to rapidly help users with tasks like calculating potential savings. This would require:

• Structured Data Interaction: Enabling the AI to accurately query and interpret transaction data, requiring a mechanism to translate natural language into precise SQL without direct integration into core banking or third-party systems.
• Regulatory Focus: Meeting stringent FINMA (Swiss Financial Market Supervisory Authority) compliance and security standards while maintaining the speed of a startup.
• Scalability for Growth: Building a scalable AI agent, capable of supporting rapid customer acquisition without proportional increases in operational overhead.
• Unified Customer View: Ensuring that the new AI agent had access to high-quality, real-time data to provide truly personalized advice.

 

Although developing the initial proof of concept for this agent appeared straightforward, bridging the gap between our prototype and a production-grade product proved to be a significant undertaking. We faced substantial challenges, including managing hallucinations, ensuring isolation, and maintaining a focused scope—all while navigating a landscape of nascent and rapidly evolving technologies. – Jérôme Ducret, Data & AI Engineer, Alpian

 

The Solution

Introducing Alpian’s AI Spending Assistant,  a conversational AI agent to provide users with high-performance, low-latency financial insights. Datatonic collaborated with Alpian to build its banking chat agent, using Google’s Agent Development Kit (ADK) and Gemini on Vertex AI. To ensure the solution is accurate and compliant, Datatonic and Alpian implemented best practices and impactful features: 

1. Google-Native “Translator” Architecture: Built using Google’s Agent Development Kit (ADK), the solution acts as a secure translator between human conversation and complex banking data. Using a Model Context Protocol (MCP) server, it converts natural language queries, like “How much did I spend on coffee?”, into precise SQL commands, allowing users to analyze spending habits instantly without the AI ever needing to directly touch the raw database.
2. Hyper-Personalized Memory Bank: Unlike standard chatbots, the agent uses Agent Engine’s Memory Bank to retain context across different sessions. This allows the assistant to remember user preferences and past interactions, evolving from a simple transactional tool into a proactive financial partner.
3. Bank-Grade Security + Guardrails: To solve the critical challenge of securing financial data within AI, the solution employs a multi-layered “Swiss Cheese” defense strategy. This includes Google Cloud Model Armor for external protection, a dedicated Input Agent to filter prohibited topics, and programmatic identity injection to ensure the AI never guesses a user ID but is strictly bound to the authenticated user data.
4. From “Black Box” to Full Observability To ensure Alpian can trust the agent 24/7, the solution moves beyond basic logging to a comprehensive observability stack. This includes real-time alerting on critical metrics (like latency spikes, HTTP errors, or cost overruns) and Google Cloud Trace to visualize the exact path of every request. 

 

Beyond these technical hurdles, Alpian and Datatonic faced a particularly tight timeline. However, seamless collaboration between the two teams drove high-velocity development, rapidly translating architectural decisions into deployed solutions. A standout feature of this effort was the use of programmatically generated artifacts, an innovative approach that allowed the LLM to query data without compromising strict isolation. – Jérôme Ducret, Data & AI Engineer, Alpian

 

The Impact

The partnership between Datatonic and Alpian successfully launched a customer-facing AI agent that supports the bank’s ambitious growth.

• Supported Alpian’s rapid growth with new AI-powered financial services capability. 
• Provided innovative new features to customers to reduce churn and help improve their customers’ decision-making.
• Optimized operational costs and unlocked more time for human advisors to focus on high-value wealth management tasks.

 

Conclusion

This new AI agent showcases Alpian’s unique value proposition that combines digital efficiency with human expertise to make private banking accessible to a wider range of customers.

Their new agent allows Alpian’s customers to quickly get insight into their spending, saving and investment decisions, which in turn allows Alpian’s advisory team to focus on answering complex questions, providing expert advice, and continuing to provide industry-leading customer service.