Custom AI training: how a dynamic companion is built and played

Catalog courses cannot cover every skill a workforce actually needs. A finance team may need AI for month-end commentary; an operations team may need AI for incident write-ups; a customer team may need AI for complaint handling. Waiting for a vendor to publish a matching course is how L&D ends up buying a generic generative-AI video and hoping people apply it. Amro Academy's dynamic companion is the product we built so organisations can commission a live AI tutor on the subject they name — then employees learn by talking, not by watching.

The create flow is deliberately short. A learner or an organisation author enters a subject or learning goal. The platform treats that as a brief, not a chat prompt: Gemini generates a structured lesson with sections, a spoken script, and knowledge checks. Audio and timing manifests are produced in the background so playback stays in sync with the text. When the companion is ready, the author is notified by email and can open it from My Journey. Individual creation uses three credits after the welcome credit; workforce programmes are quoted from headcount instead of asking L&D to multiply packs.

Play is the same voice-first lesson as a published catalog companion. Sections play as spoken tutoring. Knowledge checks pause the lesson so understanding is tested before the next block. During play the learner can ask the tutor questions by voice — Deepgram handles the live Q&A — so confusion is handled in the moment rather than in a forum two days later. Progress, quiz state and resume position are stored, which is what lets L&D measure participation instead of collecting completion ticks from a video player.

This is not a chatbot wrapper around a PDF. A dynamic companion has a curriculum skeleton, generated lesson text, TTS audio, and checks at section boundaries. That is why it can be allocated, resumed and reported like catalog training. When a companion proves useful, it can be promoted into the global catalog so other teams reuse the same tutor rather than rebuilding the topic. That promote step is how custom work becomes organisational assets rather than one person's experiment.

For L&D the commercial point is speed and fit. You do not wait a quarter for a content vendor. You do not buy a 40-hour MOOC because three modules might be relevant. You name the skill, wait for the pipeline, then put a tutor in front of the team that needs it. Pair this with OnlineTestPlus when you need a formal assessment of what people can actually do. Access to ChatGPT is not the same as AI capability — a dynamic companion is how Amro turns a named skill into structured practice.

This page is a product case study of training we created on Amro Academy, not a named-customer ROI story. We will not invent headcount, completion percentages or assessment lift. Those numbers belong on the design-partner pilot page after a paid workforce programme is measured. Book a 20-minute demo to see a dynamic companion created and played, including the progress view L&D uses to decide whether to expand the pilot.