Source Attribution API for a Major Bank

The Problem

A major enterprise bank in the US, spanning insurance and financial services, was deploying AI systems across multiple business units. As these systems became more capable, a critical question emerged: where is the AI getting its information, and can we prove that it is accurate? In regulated industries, an AI system that produces confident-sounding but unsourced answers is a liability, not an asset.

The pressure was intensifying from multiple directions. Recent AI governance legislation required financial institutions to demonstrate transparency and traceability in their AI systems. Internal compliance teams needed to audit AI-generated outputs. Business users were hesitant to trust AI recommendations without knowing the source material. And the bank's risk management framework demanded that every automated decision be traceable to its underlying data.

The bank had already invested significantly in AI capabilities, but without source attribution, adoption was stalling. Users would get an answer from an AI system and then spend time manually verifying it, sometimes defeating the purpose of having the AI in the first place. The bank needed a foundational layer that could sit beneath all of its AI systems and provide proof of where every piece of information came from.

The Solution

BetterBrain built a source attribution platform that serves as a trust layer for the bank's AI ecosystem. The platform connects to all of the bank's data sources, including document repositories, databases, internal policies, and regulatory filings. When any AI system in the bank generates a response, the source attribution API determines exactly which documents, paragraphs, and data points that response is based on.

The technical implementation uses RAG with OpenAI's gpt-4o-mini for fast attribution queries and Groq running llama-3.1-70B for bulk processing. LLM re-ranking ensures that the most relevant source documents are identified with high precision. The system does not just find vaguely related documents; it pinpoints the specific passages that support each claim in an AI-generated response.

The platform is exposed as an API that any AI system within the bank can call. This means existing AI tools do not need to be rebuilt. They simply add a source attribution step that enriches their outputs with traceable citations. The compliance team gets an audit trail, business users get confidence in the answers they receive, and the bank gets a scalable foundation for responsible AI deployment.

The Number

• Turbo-charged AI governance and compliance efforts
• Built user trust in AI systems across the organisation
• Every AI-generated answer traceable to its source document
• Compliance with emerging AI governance legislation
• Foundation layer for responsible AI deployment across all business units