Inspiration
Most runners follow generic training plans from apps like Nike Run Club or download a PDF schedule they found online. The problem? These plans are completely rigid. Life happens — you get sick, miss a week, have a bad run — and the plan just sits there, wrong and outdated. Nobody adjusts it. We wanted to build something that actually responds to how your training is going, like a real coach would.
What it does
PaceIQ is an AI running coach that generates a personalized half or full marathon training plan based on your race date, goal time, and past race history. As you log runs and tell the coach how it went in plain English — "knee felt tight at km 7" or "felt stronger than expected" — Gemini analyzes your performance and adapts your plan accordingly. After each analysis, ElevenLabs generates a personalized voice briefing so your coach can actually talk to you.
How we built it
- Backend: Python + FastAPI with SQLite for data storage
- AI: Google Gemini 2.5 for plan generation, natural language run analysis, and adaptive plan updates
- Voice: ElevenLabs text-to-speech for personalized audio coaching briefings
- Frontend: HTML, CSS, JavaScript — 4 pages: onboarding, dashboard, training plan view, and run logging
- Infrastructure: Vultr Cloud Compute for backend deployment
Challenges we ran into
- Gemini sometimes returned malformed JSON with extra text — we built a robust JSON extractor using regex to handle this reliably
- Free tier API rate limits meant careful management of Gemini calls during development
- Getting the adaptive plan logic right — prompting Gemini to understand the full context of a runner's history and make meaningful adjustments rather than just regenerating from scratch
Accomplishments that we're proud of
- The AI coaching feedback is genuinely useful and personalized — it caught that our test runner was going too fast on easy days and adjusted the plan accordingly
- ElevenLabs voice briefings make it feel like a real coaching experience, not just an app
- Built a full stack AI app solo in under 24 hours
What we learned
- Prompt engineering matters enormously — small changes to how we framed the context for Gemini dramatically changed the quality of the training plans
- Building with multiple AI APIs simultaneously (Gemini + ElevenLabs) requires careful error handling since each has its own rate limits and response formats
- A focused, well-executed idea beats a complex half-finished one every time
What's next for PaceIQ - Your AI Running Coach
- Triathlon mode — extend support to swimming and cycling alongside running
- Presage integration — contactless heart rate and recovery monitoring via camera, eliminating the need for wearables
- Deploy on Vultr GPU — fine-tune our own coaching model instead of relying on third-party APIs, enabling real-time coaching during a run with sub-second response times
- Mobile app — bring voice coaching to your phone during actual training sessions
- Social features — compare training with friends, shared coaching insights
Log in or sign up for Devpost to join the conversation.