MotoSensei

Model accuracy

Oftentimes, many who are unexperienced with automotives spend hundred, even thousands, on quick and easy repairs they can do at home. As ones who drive our 2002 and 2008 Lexus's and Nissans, it pains us heavily when we watch money get burned through such events.

As such, we built MotoSensei, an application leveraging our own trained CNN in conjunction with Gemini Vertex to assist in automotive issues and breakdowns. It helps by diagnosing mechanical failures, guiding the user through the repair process, providing the cheapest and quickest options for repair via searching for in-stock parts at nearby stores, and pointing to potentially helpful resources and videos.

In the frontend, our website application was built using JavaScript, HTML, Flask, and CSS. In the backend, Python was our language of choice, as access to TensorFlow and various other libraries proved to be extremely beneficial in building our model.

Finally, our biggest challenge was undoubtedly combining all parts of code together, as incompatibilities and difficulties quickly arose. Though individual implementations were complete, we found it difficult to run programs in parallel, unable to combine the project into one application.