The Challenge
What Ozzie Was Facing
London enterprises exploring AI assistants found that general-purpose chatbot platforms — Microsoft Copilot, generic OpenAI wrappers — could not be constrained to the data boundaries, access controls, and audit requirements demanded by regulated industries. Financial services and legal clients needed a conversational AI that would respond only from authorised knowledge bases, log every interaction for compliance purposes, and enforce role-based information access. No commercial platform offered this configuration without fundamental architectural compromise.
The Solution
What We Built
We built Ozzie as a bespoke conversational AI platform with a retrieval-augmented generation architecture that queried only designated, permission-controlled knowledge stores. Every response was logged with the source documents cited, the user identity, and a timestamp — creating a full audit trail satisfying FCA and ICO requirements. Role-based knowledge partitioning ensured that a junior analyst could not surface documents restricted to senior partners. The platform integrated with existing enterprise identity providers via SAML without requiring cloud data transfer.
Results