MrMechanic

A full size screen of the user interface.

Inspiration

One of our teammates had to go to the mechanic to fix a minor issue, which cost him more than a thousand dollars. He felt that he had to pay an absurd just because of his lack of knowledge of cars. So our team decided to develop accessible software that would help users avoid being overcharged and keep them informed about the car's condition.

What it does

It takes a very general and non-technical user input and estimates the costs to fix the car and its potential. The user input includes generally accessible information like temperature, mileage, average speed, etc. We implemented this so anybody without mechanical knowledge of cars can access our software.

How we built it

We first built a roadmap of our project and divided responsibilities. So we first found a large enough dataset to train our machine learning model through linear regression. That helped us predict costs based on information about a car. We fine-tuned Gemini AI to translate user input into quantifiable data that we could input into the machine learning module. Then we developed an accessible user interface using TypeScript, React, and Tailwind. We used Express to connect all these components.

Challenges we ran into

We spent over 4-5 hours developing a road map since we had to narrow down what we wanted to get a clear vision of our project. Then we had trouble finding appropriate datasets since the data we require is very specific. Since none of us had any previous experience with machine learning and LLMs, we weren't where to start. But through the help of mentors, we were able to make huge advancements. Then the problem was connecting all components of the project (LLM, frontend, ML), which took the bulk of our time.

Accomplishments that we're proud of

Even though none of us had any experience with LLMs and machine learning models, we were still able to train and finish a complex project. We were able to meet our original goal of helping people be informed about their vehicle's conditions.

What we learned

We learned linear regression, training/fine-tuning a large language model, connecting complex front-end and back-end through Express, and most importantly working efficiently as a team.