Notes on writing and tuning CUDA matmul kernels.
kartik-sirohi
Kartik Sirohi
IT undergrad at NIT Jalandhar · Incoming SDE Intern @ JP Morgan Chase · AI Systems
@@ research @@
P²INNs: Physio-Psychiatry Informed Neural Networks
A novel architecture for depression risk prediction integrating physiological and psychiatric indicators. Developed during a research internship at the University of Galway.
DICE-RoBERTa: Dense Image Captioning with Entropy and RoBERTa for Efficient Video Classification
Accepted & presented at the 7th International Conference on Frontiers in Computing and Systems (COMSYS 2026).
accepted & presented
The Evolution of Financial Time Series Forecasting: From Classical Econometrics to Temporal Foundation Models and Multimodal LLMs
[Preprint] A comprehensive survey and empirical evaluation across six decades of forecasting paradigms through the Capacity-Adaptability Trilemma. Investigates benchmark contamination in TSFMs/LLMs and introduces Econometrically-Informed Neural Networks (EINNs).
@@ experience @@
e8b2104
Summer 2027
UPCOMING
Software Development Engineer Intern @ JP Morgan Chase
Incoming SDE Intern for Summer 2027.
f3a9c21
Aug 2026 – present
HEAD
AI Engineering Intern @ Horrazon AI
Created web harness for agentic tasks with minimal response latency.
d4a19f2
Aug 2026 – present
AI Lead @ Google Developer Groups on Campus NIT Jalandhar
Problem setter; one of the coordinators, evaluators, and judges for MiniMol26.
7c20a5e
Aug 2026 – present
Technical Lead @ Finnest
Taught Deep Learning for Finance course as part of foundation elective; open source coordinator overseeing IndiStat (introductory project to teach student contributors open source).
1c88ae4Jan – Mar 2026
AI Research Intern @ University of Galway
Papers accepted at IEEE Access and Comsys 2026
@@ open source @@
hover or focus a node — 8 merged patches across 5 repositories
@@ projects @@index / 01–14
tinyGPT↗
mini-torch↗
linalg↗
cuda kernels↗
Onyx↗
Barnes-Hut↗
Rocky↗
Athena↗
maze-runner↗
Cortex↗
minimax↗
http-server↗
FileManager↗
Q-Toe↗
click any project for an interactive overview · or hover for quick preview
@@ posts @@
From 44 GFLOP/s to 376 GFLOP/s: Learning GPU Performance Through MatMul
Building tinyGPT in C++
Implementing a 30M-parameter GPT from scratch with NEON, SSE, and OpenMP.
Backpropagation: The Engine of Deep Learning
An intuitive walkthrough of backpropagation — from chain rule to automatic differentiation.
Intuition Behind Neural Networks
Understanding neural networks one parameter at a time.
Graph Retrieval Methods
A survey of why graph retrieval methods are useful for retrieving knowledge.
