Got my college timetable scraper working today.
Now I'm turning it into a fully automated workflow with GitHub Actions + encrypted secrets, so it can fetch updates every 24 hours without my laptop ever being on.
From “it works” to “it runs itself.” 👀
Been thinking about my voice agent a little differently.
What if every call wasn't just an interaction, but feedback?
The system could learn from real conversations and continuously improve its voice quality over time.
An AI that gets better by being used. 👀
Gave my old AI Podcaster a pretty serious upgrade today.
Multi-angle web research, source citations, and a chunked TTS pipeline that stitches everything into one audio file.
Turns out old projects are fun to revisit. 👀
Been thinking about making my AI companion feel more… alive.
Instead of waiting for me to open it, it could occasionally send a notification:
“You haven't talked to me today, idiot!”
Tiny feature, but it could make a huge difference. 👀
College absolutely ate my day today. 😭
But things are moving.
My co-founder is almost done with the dashboard, which means I'll soon be integrating it with the voice agent backend.
After that: human handoff.
The voice agent is becoming an actual product. 👀
I think I may have made a terrible decision.
I finished my voice agent for now, got some breathing room, and decided to spend that time learning how LLMs are actually built.
Now I'm seriously considering building my own.
The rabbit hole has officially begun. 💀
Finally figured out the noise cancellation for my voice agent.
After fighting distorted audio, background noise and a few cursed experiments...
BVC Telephony ended up being the way to go. 🎙️
One less problem to solve. 😭
Started with a Computer Networks lab.
Ended with me building a TCP-based remote display system with fullscreen media, a Windows keyboard hook, remote STOP control, and an .exe.
I genuinely don't know how my brain gets from A to B 😂
So we built AROM.
An AI-powered enterprise decision layer that:
• Uses RAG for contextual retrieval
• Connects multi-source data
• Recommends next steps (AUTO / REVIEW / MANUAL)
• Assigns confidence scores
Not just answers.
Actions.
At the Al Impact Summit Expo today. Clear shift in the room:
from bigger models → better system
Agents, trust, unit economics, and real-world adoption matter more than hype.
Learning a lot. Thinking even more 🚀