30+ AI agents. Onboarding to mentor.

HR teams want 1:1 mentorship for every employee. The math doesn't work — companies have thousands of people and a handful of mentors. Generic LMS content doesn't change behavior. Group coaching dilutes the signal.

Learn to Fly needed an AI-native platform that turns each employee's individual development plan (PDI) into a continuous, personalized program. Daily activities. Mentor matching. Live sessions with full lifecycle management. None of that ships from a single LLM call.

Embedded team running weekly delivery against a phase-based roadmap. Spec-driven (SpecKit), tested (Jest, Vitest, Maestro E2E), and instrumented from day one (Langfuse on every agent call). Multi-environment Terraform: dev, preview, prod. Convex preview deploys per PR.

Each employee gets a personalized program generated from their PDI. Daily activities — reflection, summary, role-play, quiz, check-in, mood prompts, content — keep momentum between mentor sessions. AnaFly coaches conversationally through the GROW framework. Human mentors get session frameworks and post-session insights. After every session, refinement questionnaires update the matching algorithm. Underneath: 30+ Mastra agents. A 69-table Convex schema with soft-delete audit trails. Zep for user context graphs. Stripe for employee-pays mentor pricing. LiveKit for live sessions.