Are you tired of just implementing simple FNNs and using prepared datasets taken
directly from Kaggle?

• Do you wish to prepare yourself for what truly a Data Science project will demand when you step into the corporate world or the numerous technical interviews before that?
• Do you wish to experience your AI model being deployed on an iOS app for on-device predictions handling products worth millions?

Then apply for the ‘Emory Pro Car Inspections: AI Vision' project right away. I was exactly in your shoes 5 years ago, as a Masters student looking for projects. Today I wish to offer you the same opportunity through my company Emory Pro.

Challenge: Leverage data from 20,000+ inspections done so far with Emory Pro to bring intelligence to vehicle inspections.

What we have: 1TB of image data - 100K images from 20K vehicles. Plus, labeled data on vehicle number, brand, year, model, damages and much more.

• Task 1: OCR - Vehicle Identification Number (VIN) Recognition in difficult conditions (snow, rain, dirt), supplemented by car image characteristics. The goal here is to develop an Optical Character Recognition (OCR) model capable of accurately identifying and extracting Vehicle Identification Numbers (VIN) from images of cars that are polluted by light reflections, dirt, rain, etc. Often the missing info can be gleaned by identifying brand, model, series, body type from vehicle photos.
• Task 2: Damage detection and severity classification: A lot of cars inspected so far have some degree of damage on them. From tiny scratches to outright missing engine bays. Our inspection reports document the damages through photos and manual labelling of the damage severity. Automating this manual step is now possible with AI vision models but differentiating a scratch from reflections is difficult. 

The tasks will utilize existing Vision models offered by Apple and Google in their mobile development kits, and to further fine tune them on AWS and supplement them with vehicle characteristics determination models. The models will be optimized for deployment on offline mobile devices (Tensorflow lite) to ensure data security, speed, offline access and privacy.
Emory Pro is deployed on AWS using a modern tech stack (Nodejs, Flutter, SQL, React, Git). Apart from vision model training/prediction, you will get to experience how a live project accumulates data and the processes required to safely expose this data for model training and inference on AWS / on-device.
You will enhance skills valued highly during technical interviews: working with IDEs, using debuggers, GIT, performance traces, AWS infrastructure deployments, SQL queries, Visualisation tools - Tableau, app Publishing on iOS/Android App Store.

Apply to this project here