The final concept is AI Buddy, a software app with three AI coaching personas, each mapped to one research barrier. The Motivator addresses low intrinsic motivation through encouraging, sentiment-aware responses. The Planner tackles time scarcity with low-friction scheduling. The Energizer adapts session intensity when users report fatigue. A shared chatbot layer called Buddy routes users to the right coach based on an Energy Check-In completed at the start of each session, the AI recommends, but the user always decides. The key design move was Tired Mode: when a user reports low energy, the app adapts their habit to a gentler version, a 20-minute run becomes a 3-minute stretch, and protects their streak regardless. Two equal-weight options ensure the user is never punished for how they feel, and every AI suggestion includes a "Why this?" link that explains the reasoning chain, directly resolving the guilt and all-or-nothing thinking that research identified as the dominant barrier.

I prototyped the concept at two levels: a medium-fidelity Figma prototype with six key screens validated through expert walkthroughs, and a functional Python chatbot using Ollama (a local LLM runtime) that tested whether the three coach personas could maintain distinct tones through system prompts alone. The chatbot runs entirely on-device, no user data leaves the machine, operationalising privacy-by-design from the prototype stage.