Blog

  • Book Your Parents a Cancer Screening

    TLDR: Cancer survival depends more on when you find it than what treatment you get. Localized cancer has 80-99% survival rates, metastatic cancer has single digits. We pour resources into new treatments while ignoring screening. If you’re in your 20s-30s, your parents are entering peak cancer risk age and probably aren’t getting screened properly. The most useful thing you can do is book them a screening package and document your family’s cancer history for yourself.

  • Your Dopamine Sources Already Decided If You'll Succeed in Life

    TLDR: Your brain releases dopamine in anticipation of reward, which means wherever you’ve trained it to fire becomes the gravitational center of all your behavior. People who get dopamine from having ideas validated optimize for being right. People who get dopamine from improving ideas optimize for getting it right. Same intelligence, completely different outcomes. Some dopamine sources compound (agency, mastery, creation, understanding) while others extract from your future self (validation, consumption, outrage, superstimulus hijacks like porn and engineered food). You cannot remove an extractive source without replacing it. Your twenties are when these patterns calcify. The honest audit: what do you reach for when you’re bored, anxious, or lonely? What do you do when no one is watching?

  • How 4 Bits Are Enough to Run Models Like ChatGPT

    This post is inspired by Paulius Micikevicius’s talk on Numerics and AI at GPU Mode. Paulius leads deep learning research at NVIDIA and authored the foundational papers on mixed precision training and FP8 formats. If you want the full depth, watch the lecture.

  • What Comes After Content Creators? Software becomes a medium.

    Every time I open Splitwise I think about how I would build it differently. The ads. The currency settings that never remember. The flows that take five taps when they should take one. I know exactly what I want. I just can’t build it.

  • Designing an optimised Audio Inference Server

    At Tensorfuse we have a lot of voice AI customers that need to host their entire pipeline on EKS. An end to end voice AI pipeline usually consists of a Speech to Text (STT) model, an LLM model for RAG or function calling, and a Text to Speech (TTS) model. There are already APIs that provide these services with closed source models like Deepgram or OpenAI’s audio models. However, when it comes to hosting open source models, there is much little information around inference optimisations and operational considerations such as scaling, routing etc.

  • It's okay to be lost in your twenties, they are meant for exploration

    When I was in my early twenties, I genuinely believed I had my life figured out. Finishing my B.Tech in just three and a half years, spending a wonderful year as a research scholar in sunny Santa Barbara, California, and preparing myself to embark on a PhD journey in computer science—everything seemed neatly laid out. Then, almost as if fate wanted to teach me something, the global pandemic struck, bringing uncertainty and forcing me to pause and reconsider.

  • 12 Rules for Life

    We all go through hard times. And honestly, adulting is more about surviving hard times than enjoying the good times. Our expectations often surpass what the universe (Him, Krishna, Parashiv, Mahavishnu - read this as whatever name suits your beliefs) has planned for us leading to turmoil both physical and mental. I have been through a lot of hard times in my life and I have learned a lot from them. Most recent one lasted for about a month and was external in nature. Things that were out of my control and I had a hard time dealing with them. Gained a bit of weight, lost a lot of sleep, had a lot of anxiety, had resultant pimples on my face and was generally not in a good place. However, I was lucky enough to have developed a set of rules for my life that never let me off the track too much.

  • God as a Programmer, Avatar as a debug process and Evil as an emergent phenomenon

    We all try to understand where we come from, where we are going, and what our purpose in life is. We look above and see the vast emptiness. We see in the past and witness vast emptiness. We look into the future and perceive infinite potential followed by a vast emptiness.

  • WTF is ASGI and WSGI in python apps?

    I’ve been working on Python-based backend development for about three years now in various forms. I primarily use Django and FastAPI, although I initially started with Flask. However, during my backend work, I frequently encountered the terms ASGI and WSGI. For example, one of my Django deployment scripts included references to asgi_app and wsgi_app, and used gunicorn to deploy these apps. Although I initially dismissed these terms as implementation details, I now find myself needing to support both ASGI and WSGI apps for my company tensorfuse. As a result, I believe it’s important to explain ASGI and WSGI to a wider audience.

  • WTF is Kubernetes Autoscaling?

    Pods and Nodes in Kubernetes

  • Why you are not seeing usage in your AWS Cost Explorer ?

    If you’re using AWS credits to cover your costs but are confused why you’re not seeing any charges in the AWS Cost Explorer, you’re not alone. The default Cost Explorer view can be misleading when credits are being applied to your account. Here’s what’s going on and how to see your true usage costs.