Neural Audio Transcription & Summarization Engine

I was looking back through some of my older projects and found this one from college.

It wasn’t for a class. I built it because I wanted a better way to go back through long lectures and team meetings without relying only on my own notes.

I still took notes normally because I definitely didn’t trust the models enough to replace that. But even with the older transcription and language models, it was useful as a second pass.

Sometimes it would catch something I missed, remind me of a point I hadn’t written down, or surface part of a conversation I wanted to revisit later.

The basic idea was simple:

Take a recording, transcribe it, and turn the transcript into something more useful than a wall of text.

What’s funny looking back is how long it took me to build.

This was before Codex, before Cursor, and before I really knew about tools like Vercel or Railway. It was mostly me, ChatGPT, VS Code, and a lot of trial and error.

I spent months testing it, fixing dependencies, and getting the whole thing running. At one point I was SSHing into IU servers just to host and run parts of it because that was the setup I knew how to use at the time.

I don’t even have access to the original deployment anymore.

But looking back, I can see the same pattern I still like building around now:

Take something messy, keep the important information, and make it easier to use.

Pretty cool seeing that show up this early.