Inspiration

Financial institutions operate in an increasingly complex environment where balance-sheet management requires professionals to analyze financial positions, investigate risks, and connect insights across multiple analytical workflows. Yet, these activities often involve fragmented tools, manual analysis, and time-consuming processes.

FRBSM (Financial Risk and Balance Sheet Management) was conceived to explore how AI can make financial intelligence more accessible, interactive, and efficient. Our vision is to bring financial analysis and AI-assisted decision-making into a unified platform.

At the center of this vision is SAM, our AI copilot—designed to provide a conversational interface for exploring financial information and interacting with analytical workflows. Rather than treating AI as just another chatbot, we want to make it a practical interface between financial professionals, their data, and the tools they use to investigate complex questions.

What it does

FRBSM is an AI-powered balance-sheet management platform designed to bring financial data, analytical capabilities, and AI-assisted workflows together in one place.

Key aspects of the platform include:

• Financial data exploration: A unified interface for viewing and exploring financial information, including legal entities and positions.
• SAM — AI copilot: A conversational interface intended to help users interact with financial information and navigate analytical workflows.
• Financial intelligence: A foundation for investigating balance-sheet positions, understanding financial risks, and supporting informed decisions.
• Workflow-oriented AI: An approach that connects conversational AI with structured analytical processes rather than relying solely on free-form text generation.
• Integrated application architecture: A web frontend, backend APIs, and database services that support the platform's financial data experience.

The broader goal is to evolve FRBSM into an agentic financial intelligence platform that helps professionals move from financial questions to structured analysis more naturally.

How we built it

We approached FRBSM as a complete financial AI application rather than an isolated chatbot.

The platform combines a web-based user interface, a FastAPI backend, PostgreSQL database services, and AI capabilities. The frontend communicates with backend APIs to retrieve financial information, while the backend provides the services and data access needed by the application.

We use Docker to containerise application services and Docker Compose to support local development. The application has also been deployed through cloud hosting platforms, including Render for the backend and Netlify for the frontend.

SAM is central to our product direction: connecting a conversational AI experience with the financial context and analytical workflows that make the platform useful.

Our development process has involved building the application incrementally, testing backend endpoints, validating database connectivity, and working through the challenges of deploying a multi-service application.

Challenges we ran into

One of our biggest challenges has been ensuring that all the components of a financial AI application work together reliably.

• Frontend-to-backend connectivity: The application works locally, but making sure the hosted frontend consistently reaches the deployed backend remains an important deployment challenge.
• Database configuration: Local development and cloud deployment require different database connection settings, and these must be configured correctly for each environment.
• Containerisation and startup: Getting the API container to start correctly, run database migrations, and listen on the appropriate port requires careful configuration.
• Reliable AI experiences: A conversational interface is only as useful as the data and analytical processes behind it. Connecting AI interactions to structured financial workflows is a core engineering challenge.
• Balancing ambition with reliability: Building an AI-powered financial platform requires more than demonstrating an impressive interface. The underlying services, data access, and analytical behaviour must work as intended.

Working through these challenges has helped us understand the importance of treating deployment, data integrity, and application reliability as fundamental parts of the product.

Accomplishments that we're proud of

We are proud to have developed a working local FRBSM application that brings together a financial interface, backend APIs, and database-backed data access.

Our accomplishments include:

• Building a web application that can display financial entities and positions.
• Establishing a FastAPI backend with health checks, database connectivity checks, and API documentation.
• Integrating PostgreSQL-backed financial data into the application.
• Creating a foundation for SAM, our AI copilot, within a broader financial intelligence platform.
• Containerising the application and establishing a cloud deployment workflow.
• Developing a product vision that connects financial-services expertise with AI-assisted reasoning and structured analytical workflows.

Most importantly, we are building toward a financial AI system where intelligence is connected to the application's data and tools, rather than confined to a standalone chat interface.

What we learned

Building FRBSM has reinforced a central lesson: useful AI products are systems, not just models.

A successful AI application depends on the interaction between models, data, APIs, infrastructure, application architecture, and user workflows. Even a capable AI model cannot deliver a reliable user experience if the underlying services are unavailable or the frontend cannot retrieve the required information.

We also learned that local functionality and production readiness are different milestones. Cloud deployment introduces additional considerations, including environment variables, database connectivity, service startup, health checks, and frontend-to-backend communication.

From a product perspective, we learned the importance of grounding AI in a specific domain. Financial professionals need more than generic answers: they need useful context, structured analysis, and outputs that can be investigated and validated.

These lessons are shaping how we approach SAM and the evolution of FRBSM.

What's next for FRBSM — AI-Powered Balance Sheet Management

Our next priority is to make the hosted FRBSM experience reliable end to end, ensuring that the frontend can retrieve live financial data from the deployed backend.

Beyond deployment reliability, our roadmap focuses on advancing SAM and strengthening the platform's financial intelligence capabilities.

We aim to:

• Advance SAM: Develop a more capable financial copilot that understands user intent and connects questions with relevant financial data and analytical workflows.
• Expand financial analysis: Build out capabilities for investigating balance-sheet positions, analyzing risk exposures, and supporting financial decision-making.
• Strengthen agentic workflows: Connect AI-assisted reasoning with structured tools and repeatable analytical processes.
• Improve reliability and validation: Emphasise dependable data access, predictable workflow execution, and verifiable analytical outputs.
• Explore accelerated AI infrastructure: Evaluate how Nebius infrastructure and NVIDIA technologies could support the platform's AI workloads where relevant.

Our long-term vision is to evolve FRBSM into an agentic AI platform for balance-sheet management and financial intelligence—helping financial professionals navigate complex analytical workflows through one unified interface, with SAM serving as their AI copilot.

We want to make financial intelligence more accessible, more interactive, and more useful—one workflow at a time.

Built With

• agentic-ai
• ai-copilot
• anthropic-claude
• docker
• docker-compose
• fastapi
• financial-services
• generative-ai
• postgresql
• python
• react
• render
• rest-api
• typescript