Gen AI platform offers future-ready capabilities

The Enterprise Risk function at a leading North American bank initiated a Generative AI (Gen AI) solution to offer a wide range of AI capabilities, including Document: Intelligence, Summarization, Generation, Translation, and more.

As the project evolved through proofs of concept and pilots, a key challenge emerged: the risk of creating a fragmented ecosystem with an overwhelming array of unmanageable bespoke solutions, model integrations, and reliance on potentially outdated models and libraries.

Based on prior engagements across clients, our team delivered thought leadership around how to develop and deliver capabilities using a platform approach. We also set up a Minimum Viable Product Team to iterate on new problem areas and solution approaches. Platform development includes generalized capabilities for:

• Setting up document ingestion pipelines, with choice of parsing approaches, embedding models and vector index stores
• A factory model along with configurations for integrating new parsers, embedding models, LLM interfaces etc., to quickly bring new capabilities to the platform
• User management, SSO integration, entitlements management
• API integration to bring in information/ data from internal and external sources
• Platform support of pgVector, Faiss, OpenSearch, Amazon Titan V2, Llama 7B, Claude Sonnet 3.5, and 3.7, etc.
• Intuitive chat interface for AI Masters - designated business users trained in Prompt Engineering and other techniques to assemble new AI/Gen AI capabilities for users through configuration - and end users

• A future-ready Gen AI platform that can easily incorporate new capabilities and updates
• Multiple specific capabilities, called skills, for use by various risk teams and business users
• A forward-looking roadmap, including ability to compose more complex capabilities using atomic capabilities