It’s that time of the year again. New Year, new resolutions, and the same old cycle we talk about how badly we failed last year, promise ourselves that this year will be different, and then somehow end up repeating the pattern. I’m talking about this because I also did something very New Year-ish: I made a weekly planner for the time I’ll be on campus. But before it sounds like another productivity stunt, I want to say one thing in my defence. ...

For the past few days, I’ve been learning something new reinforcement learning (RL) and honestly, it has been one of the most interesting shifts in perspective I’ve had while studying machine learning. During my 4th semester PRML course, I was introduced to the foundations of ML: supervised and unsupervised learning. We covered linear and logistic regression, Bayesian decision theory, decision trees, random forests, PCA, LDA, k-means, GMMs, SVMs, and later neural networks like CNNs and RNNs. ...

Yesterday, I finally completed the NVIDIA RAG (Retrieval-Augmented Generation) course, and I was genuinely intrigued by how powerful this approach is. Without fine-tuning my own LLM, I could still build a system capable of answering questions about a very specific domain accurately and reliably. That idea felt strangely familiar. I am the Vice President of the Board of Academic Interaction at my institute, and I often get asked questions that seem obvious to me. Not because people are careless, but simply because most students are unaware of the finer rules and regulations. These rules do exist on the official website (I’ll link it here) and also as PDFs but realistically, very few people read through them end to end. ...

I had started the Building RAG Agents with LLMs course by NVIDIA quite a long time ago. With the sudden boom in the field of AI and a strong urge to upskill myself, this course was always on my list. However, during the semester, things got hectic with academics, exams, fests, and other activities, and I couldn’t really give it the time it deserved. So when winter vacations finally arrived, I decided that this would be the first thing I would complete and today, I officially finished the course. ...

Over the weekend, I built another mini tool: https://apitester.krish2005.tech/. I’ve been using Postman for a while now. It’s been my go-to tool for API testing, and I’ve also experimented with other similar options like Insomnia. Truthfully speaking, these tools work incredibly well and have tons of powerful features. But here’s the thing that started bothering me. I wasn’t really using all those features they had to offer. Most of the time, I only needed the basic functionality. The advanced features came in handy occasionally, maybe when I was creating some type of documentation or generating a detailed report for APIs. But that was rare. My day-to-day goal was much simpler: I just wanted to test if an endpoint was online and working properly. When approaching things from a cybersecurity perspective, I’d check a few other quirks and edge cases. But at its core, my workflow was straightforward—sending a simple GET or POST request and checking the response. ...

A Year in Retrospect Okay, so today I’m writing about something other than tech. As this year comes to an end, I found myself looking back at everything that happened in my life over the past year. And honestly, I can say it was an overall good experience. I learned a lot, developed new skills, and started understanding human psychology a little better mostly through real-life experiences. I made new friends, reconnected with some old ones, and maybe even made a few anti-friends along the way. ...

The Problem Today was a very brief but satisfying project - building a JWT decoder and encoder from scratch. I’ve been acquainted with JWT (JSON Web Tokens) since a long time now. I use them extensively for stateless authentication in my web apps, and as a cybersecurity enthusiast, I’ve solved several CTF challenges related to JWT and some of its vulnerabilities over the years. My go-to tool for working with JWTs has always been the default debugger provided by jwt.io. It was pretty good enough with a clean interface and the option to change contents of tokens to generate new ones. But quite recently, they removed the option to edit tokens directly. Now we can only paste JWTs, decode them, and check signatures - with a completely separate tool needed to create new tokens. ...

Lately, I’ve been creating a bunch of small projects. Nothing huge, but enough that I needed a proper way to keep track of them. Sure, I could just look at the subdomain CNAME records in my DNS. But honestly — who wants to do that? And more importantly, how would I show others what I’m working on? So I started thinking about the problem, and the first thing that came to mind was Linktree. It’s literally made for this exact use case: a simple page with links to things you’re working on. ...

Lately, I felt the need to revisit web communication methods beyond the usual REST APIs. REST works great, but it’s not the only option anymore. There are other approaches like GraphQL, WebSockets, and Server-Sent Events (SSE) — each solving different problems. So instead of just reading theory, I decided to build something small and practical. That’s how https://chat.krish2005.tech came to life. The Idea I built a simple chatroom app where: ...