PRIMA is an AI-powered micro-learning assistant built for the AI Engineering stream of Telkomathon 2026, where it took 3rd place. It helps a training team turn a company document into a structured curriculum, then recommends learning items to each participant based on their skill gaps.
The curriculum builder is a guided workflow of ten steps. The author starts from a source document (PDF, DOCX, or PPTX), and PRIMA reads a short company summary from it. The author then sets the topic and target level, and the AI proposes options for the final learning objective (TLO). Once one is chosen, the AI proposes performance targets (PCS) and then enabling learning modules (ELO), of which one or more can be selected. A preview keeps the objective, performance, condition, standard, and modules aligned before the curriculum is finalized. The result shows the objective with its performance, condition, and standard, plus a learning flow with a pre-learning, in-class, and post-learning stage, each with a duration, methods, focus, and materials.
Personalization starts from the finished curriculum. For one participant or a batch, the author enters the skills to improve with the current and target level, and PRIMA returns learning items sorted by priority, each with a format and a duration. Every result is tied to the curriculum version it was generated from.
The front end is React Native with Expo Router, talking to a FastAPI backend. Azure OpenAI provides the AI layer, and PostgreSQL with pgvector handles retrieval, so learning content is matched to each user by vector similarity instead of static rules. Using pgvector inside PostgreSQL rather than a separate vector database kept the infrastructure simple on a competition timeline.
PRIMA took 3rd place in the AI Engineering stream of Telkomathon 2026.
