The speaker introduces himself and shares his passion for creating various pet projects, from Telegram bots to procedural generation games.
A childhood hobby of collecting and trading bottle caps became a way to make friends after immigrating to a new country.
The challenge of identifying duplicate bottle caps in a large collection inspired the creation of an automated tool using computer vision.
The initial approach to object detection involved simplifying bottle caps to circles and using the Hough transform algorithm, which led to issues like false positives.
Techniques like resizing, color reduction with K-means, and blob detection are used to simplify image data and refine the circle detection algorithm.
A vector database like Pinecone is used to store image features generated by the MobileNetv3 model, enabling fast and accurate similarity searches for identification.
The project was developed into a mobile app using Flutter, published to a Reddit community, and improved based on user feedback and bug reports.
The project was sunsetted due to frontend implementation challenges and a change in Firebase's billing policy, but it is still considered a valuable learning experience.
Jobs that call for the skills explored in this talk.