DEVLog - June

Hey there, how is it going?

A lot of friends have asked me why I haven’t published anything all month. The honest truth? I just wasn’t in the mood. But after a short break, I’m back and ready to share exactly what I’ve been up to this past month.

Bezier

I haven’t made much progress with Bezier yet. Implementing an offline-first architecture has been significantly harder than I anticipated. Errors are popping up everywhere, nothing seems to work, and it’s been incredibly frustrating. Fortunately, I finally got something working last week. If everything goes well from here on out, I’m aiming to release a beta sometime this month.

Follow my progress on X, and you’ll be notified before anyone else!

SignatureAPI

SignatureAPI is doing well. Dozens of weekly sign-ups are converting into paid subscriptions, which is amazing news for us. We’re providing a digital signature API designed for high volume at low costs, competing directly with giants like Adobe and DocuSign.

This month, I had to tackle a lot of tricky issues with glyph fonts. While we often take character display for granted as a trivial problem, it becomes a real pain when you’re working deep in rendering systems. The good news? Everything is working now, and you’ll see updates on the SignatureAPI blog very soon!

SignatureAPI Blog

Electronic Signature Blog and SignatureAPI News

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 demonstrate the real-world capabilities of the models we discuss in class to my students.

Deploying AI models on Cloudflare

In the first class of the month, I showed my students how to deploy AI models using Cloudflare AI Workers. We successfully deployed Whisper and a Llama 3 model.

Cloudflare AI Workers dashboard showing deployed Whisper and Llama 3 models

I also walked them through deploying models on Azure ML Studio, where we got a Microsoft Phi-3 model up and running.

Azure ML Studio showing Microsoft Phi-3 model deployment

AI friend

Next, we explored how to generate voice from text, using Coqui-TTS as our primary example. The project for the week involved building a real-time speech-to-speech chatbot, inspired by the one OpenAI demonstrated just the week before.

Real-time speech-to-speech AI chatbot demo using Coqui-TTS

Landing a Rover on the Moon

Moving on to Reinforcement Learning, I took the opportunity to introduce my students to the world of Gymnasium and Stable Baselines. The challenge for the week? Training an AI agent to land a rover on the moon!

Lunar Lander AI trained with Gymnasium and Stable Baselines

Gymnasium Workshop

The final week was an assessment of my students’ skills. I wanted to see if they could identify the core elements of a reinforcement learning problem and apply them to solve various Gymnasium environments. We saw a little bit of everything!

Car Racing environment solved with reinforcement learning

Bipedal Walker environment with trained AI agent

Acrobot environment with reinforcement learning solution

Mountain Car environment solved by students

Some of the problems resolved by my students…

The Next Steps

This month was awesome, and I’m expecting even more from July. With the semester coming to an end, I’ll have more time to dedicate to other projects and experiences. I plan to document it all right here on the blog, so if you want to stay updated, be sure to join the newsletter!

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.