GoMile was my team's winning entry to the BUILD IRL hackathon, which was run by Solana and the Trinity Claude Builder's Club at Dogpatch Labs in Dublin. When coming up with an idea, we realised that, for self-driving cars to expand outside large cities, companies will have to collect and label large swathes of driving data.
To help address this, we created GoMile, a webcam that allows drivers to video themselves as they drive on a daily basis. This data would then be sent to our servers, where it would be labelled, categorised and sold for training. In return, drivers would get paid in our Solana token, MILE, ensuring transparency and a fair market rate. We also gave drivers a bonus for exploring new areas.
Our recording setup involved a webcam connected to a Raspberry Pi with a Sense HAT, allowing us to gain more in-depth data. To label what we collected, we used a YOLO model which ran on my MacBook. We then used Qwen to identify things such as crashes or near misses, as these can raise the value of the clips for training. If I was doing it again, I would probably use a more complex setup for this, which would have allowed us to use smaller and faster models.
My main role was in setting up the payouts for drivers and our landing page. The project was finished in around six hours, and we were able to successfully record training data (by cycling around Dublin), label it and pay ourselves in MILE using the Solana devnet. We took home 1,500 USDC as a team.