TriFit

Logo
Home Screen
Recovery Screen
Recipe example
Progress page
Milestones
Calender
Chatbot
Profile page

Inspiration

Training for a race, whether its a marathon, a Hyrox or even a triathlon can be intimidating especially as a beginner since most of them don't know where to start. Private coaches cost anywhere between $150-$400 per month. At that price most people aren't willing to pay, they would rather just wing it, resulting in a pretty severe risk of injury. We wanted to build the thing a real coach gives you that no tracking app does: a plan that adapts to your life in real time, helping beginners get into the marathon and track community.

What it does

TriFit is an AI training coach built especially for beginners. After signing up users are required to fill out a goal along with information about their fitness that helps us determine the best fit plan for them to reach their goal. We don't just set a goal, we instead help reach that goal with a personalized agent, and regular check-ins to monitor your progress. We know things come up, that's why our program is dynamic. It adapts your week to your schedule even if it's a sudden issue.

How we built it

Front-end: React Native
Back-end: FastAPI and Python
Database: TigerData
LLM: Custom LLM for plan generation
AI: Gemini AI for chatbot

Challenges we ran into

Over the course of the project we had 3 big hurdles, those three being front-end design. We wanted a modern feel that was cool but also not too boring, but we also wanted a design that would fit the feel of the app. The next one was the LLM, specifically integrating it and having it generate the exact type of data we needed to be able to display it properly for the users. We settled for a json file that generates a month's plan and adjusts it as the user needs. Finally the issues that took us the longest was voice integration for Jim (ai agent), both to have him talk and being able to talk to him.

Accomplishments that we're proud of

Shipping a working end-to-end flow (onboarding -> logging -> adaptive plan -> dashboard) in a single hackathon Designing a gamification system that motivates the user through self competition Building a system that actually adapts to real life input like missed session or reported soreness and not just a static plan

What we learned

We learned a lot about context for AI specifically since we had to give the LLM all the fitness information about the user when having to revise the users plan due to changes. We also gained valuable insights into the endurance training domain, specifically how to translate complex, professional coaching methodologies into an approachable, beginner-friendly application architecture.

What's next for TriFit

Watch Integration: We plan to implement apple watch and galaxy watches into the system to help simplify tracking
Fine-tune the LLM: While our LLM is functional it is far from perfect, we plan to improve upon it a lot more and make it much more accurate
Map View: We believe implementing a map that shows your runs would be an interesting and fun addition