Adithya Kamath

QR-decomposition-kernel. [GPU MODE Contest] qr decomposition on blackwell GPUs

dagu. Local-first workflow engine with a Web UI for small teams. Define DAGs in a declarative YAML format. Self-contained and no DBMS required. Use any AI agent to manage your DAGs.

oxydra. oxydra - An agent with many heads, zero overhead

kvm-hello-world. A minimal kvm example

sglang-jax. JAX backend for SGL

nano-vllm. Nano vLLM

ares. Agentic Research and Evaluation Suite

neu. C++

mini-sglang. A compact implementation of SGLang, designed to demystify the complexities of modern LLM serving systems.

Emerging-Optimizers. Python

hardened_malloc. Hardened allocator designed for modern systems. It has integration into Android's Bionic libc and can be used externally with musl and glibc as a dynamic library for use on other Linux-based platforms. It will gain more portability/integration over time.

gemlite. Fast low-bit matmul kernels in Triton

torchtitan. A PyTorch native platform for training generative AI models

grain. Library for reading and processing ML training data.

picotron_tutorial. Python

autofsdp. Fully-sharded data parallelism (FSDP) in jax with minimal code changes.

kvax. A FlashAttention implementation for JAX with support for efficient document mask computation and context parallelism.

flite. A small fast portable speech synthesis system

levanter. Legible, Scalable, Reproducible Foundation Models with Named Tensors and Jax

haliax. Named Tensors for Legible Deep Learning in JAX

Tsukasa-Speech. a Frontier Japanese Speech Generation net

ms-swift. Use PEFT or Full-parameter to CPT/SFT/DPO/GRPO 600+ LLMs (Qwen3.6, DeepSeek-V4, GLM-5.1, InternLM3, Llama4, ...) and 300+ MLLMs (Qwen3-VL, Qwen3-Omni, InternVL3.5, Ovis2.5, GLM4.5v, Gemma4, Llava, Phi4, ...) (AAAI 2025).

multi-latent-attention. Python

vui. Real-time voice assistant — WebRTC streaming, faster-whisper ASR, local LLM, Vui Nano (300M) TTS. OpenAI Realtime API compatible. Voice cloning, barge-in, ~9× realtime on a 4090. Apache 2.0.

atropos. Atropos is a Language Model Reinforcement Learning Environments framework for collecting and evaluating LLM trajectories through diverse environments

verifiers. Our library for RL environments + evals

avataRL. rl from zero pretrain, can it be done? yes.

SuperTokenizer. Multi-Word Probabilistic based supertokenizer

NanoPoor. NanoGPT-speedrunning for the poor T4 enjoyers

Paper-Replications. A repository consisting of paper/architecture replications of classic/SOTA AI/ML papers in pytorch

tokenizers-mbpe. Morphologically biased byte-pair encoding pre-tokenization

ray-tpu. Python

maxtext. A simple, performant, and scalable Jax LLM!

torchprime. torchprime is a reference model implementation for PyTorch on TPU.

open-infra-index. Production-tested AI infrastructure tools for efficient AGI development and community-driven innovation

automq. Diskless Kafka® on S3. 10x Cost-Effective. No Cross-AZ Traffic Cost. Autoscale in seconds. Single-digit ms latency. Multi-AZ Availability.

xla. Enabling PyTorch on XLA Devices (e.g. Google TPU)

torchacc. PyTorch distributed training acceleration framework

FlashModels. Fast and easy distributed model training examples.

tokenadapt. 0-Shot Tokenizer Transplant

typescript-go. Staging repo for development of native port of TypeScript

picotron. Minimalistic 4D-parallelism distributed training framework for education purpose

FlashMLA. FlashMLA: Efficient Multi-head Latent Attention Kernels

ohara. Collection of autoregressive model implementation

naptha-sdk. Naptha is a framework and infrastructure for developing and running multi-agent systems at scale with heterogeneous models, architectures and data

naptha-node. Software for running personal agents locally that can interact with personal agents on other devices

DeepSpeed. DeepSpeed is a deep learning optimization library that makes distributed training and inference easy, efficient, and effective.

tensorzero. TensorZero is an open-source LLMOps platform that unifies an LLM gateway, observability, evaluation, optimization, and experimentation.

advent-of-code. Rust

NextTrade. A system that performs algorithmic trading

omniparse. Ingest, parse, and optimize any data format ➡️ from documents to multimedia ➡️ for enhanced compatibility with GenAI frameworks

marker-api. Easily deployable 🚀 API to convert PDF to markdown quickly with high accuracy.

dungbeetle. A distributed job server built specifically for queuing and executing heavy SQL read jobs asynchronously. Separate out reporting layer from apps. MySQL, Postgres, ClickHouse.

kaf-relay. Replicate and sync Kafka topics between clusters in realtime. Supports topic re-mapping, healthchecks, and hot failovers for high availability.

smolar. a tiny multidimensional array implementation in C similar to numpy, but only one file.

valet. Job queuing service and async task runner.

low-level-design-primer. Dedicated Resources for the Low-Level System Design. Learn how to design and implement large-scale systems. Prep for the system design interview.

LLD. Curated Collection of all Low level design Questions and implementation asked in major Tech companies , Get yourself prepared for the LLD round and ace the interview.

pg_timeseries. Simple and focused time-series tables for PostgreSQL

postgres-page-inspector. Postgres page inspector

toxicVault. A distributed datastore that never goes down.

goharvest. Transactional outbox harvester for Postgres → Kafka, written in Go

go-food-delivery-microservices. 🍕 A practical food delivery microservices, built with Golang, Domain-Driven Design, CQRS, Event Sourcing, Vertical Slice Architecture, Event-Driven Architecture, and the latest technologies.

diztl. Share, discover & download files in your network 💥

rill. Go toolkit for clean, composable, channel-based concurrency

gokatas. Internalize Go programming techniques

borgo. Borgo is a statically typed language that compiles to Go.

go-coffeeshop. ☕ A practical event-driven microservices demo built with Golang. Nomad, Consul Connect, Vault, and Terraform for deployment

radish. Super fast drop-in replacement of the in memory key-value store Redis, made in Golang

tiny-gpu. A minimal GPU design in Verilog to learn how GPUs work from the ground up

MIT-6.824. Basic Sources for MIT 6.824 Distributed Systems Class

6.824. Labs of MIT 6.824: Distributed Systems.

MIT-6.824-Lab. 💡MIT 6.824: Distributed Systems💡 is a core 12-unit graduate subject. It will present abstractions and implementation techniques for engineering distributed systems. Major topics include fault tolerance, replication, and consistency.

mit6.824. MIT6.824 Distributed Systems

agungdpcms. Backend for my cms powered by GraphQL and Websocket

reverse-proxy. Repository for manage all my API's

gojek-parking-lot. GO-JEK Challenge

timeq. A fast file-based priority queue

muffet. Fast website link checker in Go

ShortNews. Apply LLM and TTS for a rss news reader

distributedfilesystemgo. distributedfilesystemgo

jocko. Kafka implemented in Golang with built-in coordination (No ZK dep, single binary install, Cloud Native)

awesome-distributed-systems. A curated list to learn about distributed systems

project-layout. Standard Go Project Layout

Go-gPRC-microservices-Docker-example. Go, gRPC, microservices, docker example

open-gpu-kernel-modules. NVIDIA Linux open GPU with P2P support

gossip-glomers. My solutions to the Glomers Challenge: a series of distributed systems challenges.

vitess. Vitess is a database clustering system for horizontal scaling of MySQL.

badger. Fast key-value DB in Go.

wtf. WTF Dial is an example web application written in Go.

blaze. Simple and minimal Go template for building fast and mantainable HTTP services

ragflow. RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context layer for LLMs

llm.c. LLM training in simple, raw C/CUDA

computer-science. 🎓 Path to a free self-taught education in Computer Science!

SWE-agent. SWE-agent takes a GitHub issue and tries to automatically fix it, using your LM of choice. It can also be employed for offensive cybersecurity or competitive coding challenges. [NeurIPS 2024]

liquibase-demo. Kotlin

data-genie. A synthetic data generation pipeline using LLMs and websearch over a curriculum

Devon. Devon: An open-source pair programmer

Sequoia. scalable and robust tree-based speculative decoding algorithm

guide. The Uber Go Style Guide.