AI Mentor - SOP Training System
A training system, not a document chatbot: SOPs become assessment, every wrong answer is tracked, and review is scheduled per person by FSRS
Overview
The difference between AI Mentor and a retrieval chatbot is what happens after the answer. A chatbot closes the conversation and the knowledge leaves with it; AI Mentor records that this person, on this topic, was wrong, and puts the fact back in front of them before they lose it. It indexes a company's standard operating procedures, then teaches from them.
Documents are split into knowledge items and embedded into pgvector so the mentor answers questions with the source paragraph attached rather than a plausible guess. Quiz questions and flashcards are generated from the same material, scheduled for review by FSRS so each person revisits a fact on the day they are about to forget it.
A practice mode plays the difficult customer and scores the handling against the procedure. Managers get a forgetting curve per topic and a gap report showing which SOPs the team consistently gets wrong - usually a sign the document is unclear rather than that the staff is careless. The LLM provider is configuration, so procedures that must not leave the network can be served by a self-hosted Ollama or vLLM endpoint.
Interactive demo
One card, four learners, four different schedules
Rate how well you recalled each card and watch the schedule recompute. This is where AI Mentor differs from a chatbot: it does not only answer, it decides when you have to meet the fact again.
A customer asks for a refund on day 9 after delivery. What do you do?
The scheduling formula here is a simplified illustration. The real system uses FSRS with parameters fitted to each person's actual review history. The SOP text is an example, not any customer's procedure.