DEVLog - May

I can’t believe it’s already June. Time is flying, so let’s take a moment to review everything I tackled in May.

Bezier

If you follow me on social media, you might have noticed I’ve been a bit quieter than usual with updates.

The reason? For most of the month, the app has been in pieces. I’ve been heads-down implementing a true offline-first architecture. This required rewriting a massive amount of backend and internal logic, which effectively broke the application. My entire focus this month has been rebuilding it to get back to where we were at the start of May—but now with offline superpowers.

SignatureAPI

SignatureAPI has seen solid growth over the last few weeks, onboarding dozens of new users weekly since we deployed the Microsoft Power Automate connector.

This month, we’re planning to release a much-requested internationalization feature and polish several key user flows.

If you’re looking for a cost-effective, high-volume digital signature API, definitely check out SignatureAPI.

Electronic Signature API — SignatureAPI

Electronic signature API for workflows, applications and platforms

signatureapi.com

AI and ML

As many of you know, I’m a professor of Artificial Intelligence and Machine Learning. This month, I implemented several projects to help my students understand the real-world capabilities of the models we discuss.

RemoveBG

The first project involved removing image backgrounds using the Segment Anything Model (SAM). We identified image masks in real-time to cleanly separate the subject from the background.

RemoveBG demo - real-time background removal using Segment Anything Model

Image Upscaler

Continuing with autoencoders, I demonstrated how to leverage existing models on Hugging Face and Replicate, using image upscaling and super-resolution as our primary use case.

Image upscaler demo using autoencoders for super resolution

Denoiser

Moving into audio processing, I showed my students how to build an audio denoiser tool using SepFormer in the time domain.

Audio denoiser demo using SepFormer in the time domain

Trader

To introduce Recurrent Neural Networks, we built a trading bot to predict Bitcoin prices using the Yahoo Finance API. We compared Simple RNNs, LSTMs, and GRUs to see which performed best. Our model predicted BTC would hit $72,000 USD by mid-June. Let’s see if the AI knows something we don’t!

Bitcoin price prediction chart showing RNN, LSTM, and GRU model performance

Translator

The final project was implementing a translator app using our own transformer from scratch. This helped students understand the inner workings of the architecture. The result? An app that translates Spanish to English with roughly A1-level proficiency.

Spanish-English translator demo built with a custom transformer model

The Next Steps

This month was wonderful, and June promises even more awesome projects. I’ve been swamped, so blog posts have been sparse, but I expect to carve out some time in the next few weeks. Expect new posts on offline-first architecture, AI/ML applications, and other updates from my journey. See you then!

If you enjoy the content, please don’t hesitate to subscribe and leave a comment! I would love to connect with you and hear your thoughts on the topics I cover.