<feed xmlns="http://www.w3.org/2005/Atom"> <id>https://karthikbalasubramanian.github.io/</id><title>Karthik Balasubramanian</title><subtitle>Karthik Balasubramanian writes about LLM inference engineering and GPU kernels: Triton and CUDA, attention, KV cache, vLLM and SGLang serving, benchmarking and observability.</subtitle> <updated>2026-11-19T16:52:27-08:00</updated> <author> <name>Karthik Balasubramanian</name> <uri>https://karthikbalasubramanian.github.io/</uri> </author><link rel="self" type="application/atom+xml" href="https://karthikbalasubramanian.github.io/feed.xml"/><link rel="alternate" type="text/html" hreflang="en" href="https://karthikbalasubramanian.github.io/"/> <generator uri="https://jekyllrb.com/" version="4.4.1">Jekyll</generator> <rights> © 2026 Karthik Balasubramanian </rights> <icon>/assets/img/favicons/favicon.ico</icon> <logo>/assets/img/favicons/favicon-96x96.png</logo> <entry><title>The Flashlight and the Shadow: How Dot Products and Normalization Power AI Attention</title><link href="https://karthikbalasubramanian.github.io/ml/2026/09/02/utility_of_dot_products_in_attention.html" rel="alternate" type="text/html" title="The Flashlight and the Shadow: How Dot Products and Normalization Power AI Attention" /><published>2026-09-02T00:00:00-07:00</published> <updated>2026-11-19T12:21:51-08:00</updated> <id>https://karthikbalasubramanian.github.io/ml/2026/09/02/utility_of_dot_products_in_attention.html</id> <content type="text/html" src="https://karthikbalasubramanian.github.io/ml/2026/09/02/utility_of_dot_products_in_attention.html" /> <author> <name>Karthik Balasubramanian</name> </author> <category term="LLM Foundations" /> <summary>From flashlight shadows to multi-dimensional Transformer attention: an intuitive, first-principles deep dive into dot products, matrix operations, and why self-attention works the way it does.</summary> </entry> <entry><title>The Great AI Compromise: How Autonomous Agents Will Reclaim Our Idling GPUs</title><link href="https://karthikbalasubramanian.github.io/ml/2026/08/26/the_great_ai_compromise.html" rel="alternate" type="text/html" title="The Great AI Compromise: How Autonomous Agents Will Reclaim Our Idling GPUs" /><published>2026-08-26T00:00:00-07:00</published> <updated>2026-11-19T15:18:07-08:00</updated> <id>https://karthikbalasubramanian.github.io/ml/2026/08/26/the_great_ai_compromise.html</id> <content type="text/html" src="https://karthikbalasubramanian.github.io/ml/2026/08/26/the_great_ai_compromise.html" /> <author> <name>Karthik Balasubramanian</name> </author> <category term="Inference Engineering" /> <summary>An exploration of FLOPs, memory bandwidth, MFU across GPT-2 to Claude 3.5 Sonnet, and how long-horizon autonomous agents will unlock maximum GPU throughput.</summary> </entry> <entry><title>Facial Emotions Recognition using Heuristic Learning and Transfer Learning</title><link href="https://karthikbalasubramanian.github.io/ml/2019/05/17/Karthik-_FER_Capstone_Report.html" rel="alternate" type="text/html" title="Facial Emotions Recognition using Heuristic Learning and Transfer Learning" /><published>2019-05-17T00:00:00-07:00</published> <updated>2026-11-19T15:52:59-08:00</updated> <id>https://karthikbalasubramanian.github.io/ml/2019/05/17/Karthik-_FER_Capstone_Report.html</id> <content type="text/html" src="https://karthikbalasubramanian.github.io/ml/2019/05/17/Karthik-_FER_Capstone_Report.html" /> <author> <name>Karthik Balasubramanian</name> </author> <category term="Deep Learning" /> <summary>Emotion Recognition and Pattern Recognition</summary> </entry> <entry><title>How to Train an agent to play the game of Black Jack?</title><link href="https://karthikbalasubramanian.github.io/ml/2019/03/29/Monte_Carlo.html" rel="alternate" type="text/html" title="How to Train an agent to play the game of Black Jack?" /><published>2019-03-29T00:00:00-07:00</published> <updated>2026-11-19T15:52:59-08:00</updated> <id>https://karthikbalasubramanian.github.io/ml/2019/03/29/Monte_Carlo.html</id> <content type="text/html" src="https://karthikbalasubramanian.github.io/ml/2019/03/29/Monte_Carlo.html" /> <author> <name>Karthik Balasubramanian</name> </author> <category term="Reinforcement Learning" /> <summary>Reinforcement learning using Monte Carlo Methods</summary> </entry> <entry><title>Dynamic Programming - An easier Reinforcement learning setup</title><link href="https://karthikbalasubramanian.github.io/ml/2019/03/15/dp.html" rel="alternate" type="text/html" title="Dynamic Programming - An easier Reinforcement learning setup" /><published>2019-03-15T00:00:00-07:00</published> <updated>2026-11-19T15:52:59-08:00</updated> <id>https://karthikbalasubramanian.github.io/ml/2019/03/15/dp.html</id> <content type="text/html" src="https://karthikbalasubramanian.github.io/ml/2019/03/15/dp.html" /> <author> <name>Karthik Balasubramanian</name> </author> <category term="Reinforcement Learning" /> <summary>Solve a classic MDP ( Exploring Frozen Lake) using Dynamic Programming</summary> </entry> </feed>
