Prophet Profiling

Challenge

Prophet Profiling's system is explained in a handbook that runs to hundreds of pages. It covers how the profiling model works and how to apply it, and it serves three different audiences: coaches, HR professionals and the people being profiled. Each of them needs a different slice of it, and almost none of them were reading it.

Instead, they contacted the Prophet team directly. The knowledge existed and was thorough, but the format was wrong for the way people actually use it. Prophet asked us for a way to make that knowledge accessible at the moment a question comes up, so users could find their own answers rather than joining a support queue.

Process

We built a custom RAG database containing the full manual, so the assistant answers from Prophet's own documentation rather than general knowledge. Every response is grounded in the handbook and consistent with the methodology. For the questions the Prophet team fields most often, we authored specific answers and prioritised them in retrieval.

Prophet's data compliance requirements meant this could not run on off the shelf AI services. We built it on Amazon Bedrock within their own AWS account, so the manual and every conversation stay inside Prophet's environment, are never used to train third party models, and remain under their control.

Much of the handbook is not text. It explains the model through comparison tables, scales and diagrams, so we gave the assistant the ability to generate tables, charts and other rich media when the answer calls for it. A user asking how two profile types differ gets a table rather than three paragraphs.

Solution

The chat interface sits inside the Prophet platform, where users already are. Coaches, HR professionals and profiled users each ask in their own words and get an answer drawn from the handbook, formatted in whatever way suits the question.

The handbook has not been reduced or rewritten. It is now queryable, and the knowledge inside it reaches people who were never going to read it end to end. Routine questions resolve in the platform, and the Prophet team spends its time on the conversations that need a person.