Back on Track After a Few Weeks

Getting Back Into the Rhythm

So, as you guys know, I'm writing a blog after a long time -- almost like 2-3 weeks. Yeah, I was just a bit confused all these days about what I'm doing, whether it's good for me and all. But now that I'm back on track, let's discuss what has happened over these few weeks.

Current Work: Hospital Management System

I've mostly been working on a hospital management system where I'm currently employed. I primarily use VSCode with Copilot powered by either Sonnet-4 or GPT-4.1, and that combination has been going pretty well.

My workflow looks like this:

1. Create a perfect set of tasks with each line clearly defining the logic I need to add
2. Write a bit of logic before implementing the full solution
3. Tell Copilot to complete it based on my setup

I know all the purists are like, "Oh, why would you do that? You're not writing code!" But here's the thing--95% of the code online is already written by AI. Why the hell do people care if I use Copilot? I'm working with React and Django, and for most React-based work, Copilot is more than enough to get the job done efficiently.

The Cool Stuff: TinyLLaMA Implementation

Now for the exciting part! I'm working on implementing TinyLLaMA models (created by Andrej Karpathy) into various projects.

Initially, I looked into implementing it for the web, but I'm dumb and wasn't able to make much progress there. After that setback, I finally planned to implement it into my Romanji challenge project -- something a lot of recruiters ask about during interviews and how I have implemented it in.

I thought: how cool would it be to use a small 256k parameter LLM running right from the browser while being deployed for free on sites like Netlify or Vercel? The idea of having AI-powered functionality without any server costs is pretty appealing.

Another project I've planned but haven't started is the SteamDeck plugin I've been working on for a while. But it's just way too much work to test properly, so once I'm a bit more free, I'll try to work on it again.

Cool Things I Discovered

Here are some fascinating discoveries from the past few weeks but I only remember about the ones I saw the past week:

ESP32 LLM Implementation

ESP32-LLM by DaveBben: A project to run a Large Language Model (LLM) directly on the ESP32! Unfortunately, I had a replica ESP32 board without PSRAM, so I wasn't able to test it. However, I did get to explore ESP-IDF and learn how to use it, which was pretty cool.

APKLab for Mobile Analysis

APKLab: This tool seamlessly integrates the best open-source tools like Quark-Engine, Apktool, Jadx, uber-apk-signer, and apk-mitm into VS Code. Now you can focus on app analysis and get it done without leaving your IDE.

Velox Amphibious Robot

I came across this amazing video of an amphibious robot named Velox, developed by Pliant Energy Systems. It mimics the locomotion of various sea creatures, allowing it to navigate diverse environments like water, ice, and land. The engineering behind biomimetic robotics is absolutely fascinating.

Anthropic's Defense Contract

Anthropic's partnership with the Department of Defense: Anthropic's $200M contract with the U.S. Department of Defense marks a significant shift from its AI safety mission, aligning with a 2023 RAND Corporation study showing 70% of AI firms pivot to defense contracts under national security pressures.

This move coincides with reports revealing similar $200M deals with OpenAI, Google, and xAI, reflecting a broader Pentagon strategy to counter China's AI advancements. It's interesting to see how the AI landscape is evolving, especially when companies that started with safety-first missions are adapting to geopolitical realities.

Moving Forward with Purpose

Let's end on a better note -- I'll try to be consistent and make improvements!

The past few weeks taught me that taking breaks is sometimes necessary for clarity, but I'm excited to get back into regular blogging. There's so much happening in the tech world, from tiny LLMs running on microcontrollers to major AI companies reshaping their strategies.

Stay tuned for more updates on the TinyLLaMA implementation, the SteamDeck plugin progress, and whatever other interesting rabbit holes I end up exploring. The goal is consistency, improvement, and sharing the journey with all of you.

Until next time, keep building cool stuff!

ps: just testing out a new custom rehype plugin to detect ascii