Inspiration
Personal training and specialized nutrition are a privilege few can afford. For many beginners, going to the gym without guidance leads to severe injuries, burnout, and quitting. We wanted to build a "Social Good" solution: an elite, empathetic AI coach that democratizes safe, science-based fitness by adapting to the user's real-time physical and mental fatigue.
What it does
AuraFit is an Agentic AI fitness ecosystem. Instead of static plans, Aura uses daily check-ins to gauge readiness. If it detects severe leg fatigue or poor sleep, the AI dynamically adjusts the day's macros (e.g., increasing carbs for recovery) and safely rewrites the workout plan (switching to upper-body or active recovery), explaining the biomechanical why behind every change to educate the user.
How we built it
We built the backend using Laravel to handle our robust database and API routes, paired with a sleek, interactive frontend using React, TypeScript, and TailwindCSS. The core brain connects to Large Language Models using strict JSON-mode prompt engineering to ensure safe, parseable workout structures. We also built an internal Admin Dashboard to track AI token usage in real-time, ensuring our business model is hyper-profitable and scalable.
Challenges we ran into
Our biggest hurdle was the "Token Eater": Image Processing. When we integrated our macro-scanning feature, high-res photos from modern phones consumed up to 26,000 tokens per request. We solved this by implementing an aggressive frontend image compression pipeline and optimizing our prompts, reducing token consumption by 80% and bringing our operational cost down to mere fractions of a cent per user.
Accomplishments that we're proud of
- Building a truly "Agentic" AI that doesn't just chat, but actively triggers state changes in the UI (updating progress bars and databases autonomously).
- Developing a commercial-grade, cyberpunk-inspired UI/UX from scratch.
- Creating a financially viable SaaS model validated by our custom Admin metrics dashboard.
What we learned
We learned the critical importance of Prompt Engineering in production. By injecting strict rules about workout splits (Push/Pull/Legs) and preventing "hallucinated" exercise combinations, we learned how to constrain an LLM to produce safe, predictable, and highly structured JSON data.
What's next for AuraFit
We plan to launch AuraFit as a Progressive Web App (PWA).
Log in or sign up for Devpost to join the conversation.