Projects Recently Starred by Experts

Starred 13 hours ago by

Pawel Garbacki(Cofounder of Fireworks AI). Also starred by

Chip Huyen(Author of "AI Engineering", "Designing Machine Learning Systems"),

Elie Bursztein(Cybersecurity Lead at Google DeepMind),

Jeff Hammerbacher(Cofounder of Cloudera), and

9 more.

RouteLLM by lm-sys

1.0%

Framework for LLM routing and cost reduction (research paper)

Drop-in OpenAI client replacement; routes queries to cheaper models.
Trained routers reduce costs up to 85% while maintaining GPT-4 quality.
Supports local models; calibrates cost/quality thresholds.
Includes evaluation framework for comparing router performance on benchmarks.

Created 2 years ago

Updated 1 year ago

Starred 18 hours ago by

Travis Fischer(Founder of Agentic).

shieldcn by jal-co

7.2%

539

Open-source badge service rendering UI components as SVGs

Generates visually appealing badges styled like shadcn/ui components using Satori.
Supports numerous providers (npm, GitHub, Docker) and custom dynamic JSON endpoints.
Offers a CLI tool for easy badge generation and integration into READMEs.
Provides self-hosting options and an agent skill for automated badge management.

Created 2 months ago

Updated 12 hours ago

Starred 1 day ago by

Elvis Saravia(Founder of DAIR.AI). Also starred by

Travis Fischer(Founder of Agentic).

birdclaw by steipete

9.1%

Local Twitter memory and agent data store

Archive and live tweet data locally in SQLite for agent access and automation.
Provides a web UI and CLI for triage, search, and AI-assisted inbox scoring of mentions and DMs.
Git-friendly text backups and stale-aware reads ensure data integrity and efficient local state management.
Local-first design with no auth layer ensures privacy and offline functionality for tweet archiving and interaction.

Created 3 months ago

Updated 2 days ago

Starred 1 day ago by

Jonathan Ragan-Kelley(Professor at MIT).

mcpc by apify

1.4%

702

Universal CLI for Model Context Protocol interactions

Interacts with MCP servers via HTTP/stdio, managing persistent sessions.
Powers AI agents with JSON code mode and secure sandboxing via MCP proxy.
Features OAuth 2.1 authentication and OS keychain for secure credential management.
Supports tool calling, resource access, and schema validation for robust integrations.

Created 6 months ago

Updated 6 days ago

Starred 1 day ago by

Jesse Clark(Cofounder of Marqo). Also starred by

Amin Ahmad(Cofounder of Vectara),

Woosuk Kwon(Coauthor of vLLM),

Jeff Hammerbacher(Cofounder of Cloudera), and

17 more.

verl by verl-project

0.5%

22k

RL training library for LLMs

Flexible RL algorithms via hybrid-controller programming model.
Seamless LLM infra integration with FSDP, Megatron-LM, vLLM, SGLang.
Efficient actor model resharding with 3D-HybridEngine.
Supports PPO, GRPO, ReMax, REINFORCE++, RLOO, PRIME, DAPO, DrGRPO.

Created 1 year ago

Updated 1 day ago

Starred 1 day ago by

Eric Zhu(Coauthor of AutoGen; Research Scientist at Microsoft Research). Also starred by

Eric Zhang(Founding Engineer at Modal),

George Hotz(Author of tinygrad; Founder of the tiny corp, comma.ai),

Jeremy Howard(Cofounder of fast.ai), and

11 more.

slime by THUDM

4.1%

LLM post-training framework for RL scaling

Enables efficient training via Megatron integration with SGLang.
Supports flexible data generation workflows via custom interfaces.
Includes HF checkpoint conversion to Megatron's torch_dist format.

Created 1 year ago

Updated 1 day ago

Starred 1 day ago by

Kent Dodds(Cofounder of Remix). Also starred by

Steve Sewell(Founder of Builder.io).

agent-native by BuilderIO

25.9%

Build agent-native applications with deeply integrated UIs and autonomous agents

Seamlessly connect UIs and agents, sharing state and database for bidirectional control and context awareness.
Enables self-improving applications where agents can modify features, fix bugs, and refine the UI over time.
Develop production-grade SaaS apps from cloneable templates, offering full ownership and customization without vendor lock-in.
Supports diverse AI models, any SQL database, and agent-to-agent communication for flexible, extensible systems.

Created 3 months ago

Updated 23 hours ago

Starred 2 days ago by

Pawel Garbacki(Cofounder of Fireworks AI). Also starred by

Dan Abramov(Core Contributor to React; Coauthor of Redux, Create React App) and

Jared Palmer(SVP at GitHub; Founder of Turborepo; Author of Formik, TSDX).

next-evals-oss by vercel

1.5%

276

AI agentic framework for Next.js coding evaluations

Tests AI models on generating and refactoring Next.js codebases.
Employs structured prompts and Vitest assertions within self-contained Next.js projects.
Automated runner efficiently manages new evals and models, skipping completed tasks.
Covers diverse Next.js features: routing, data fetching, server actions, and middleware.

Created 8 months ago

Updated 1 day ago

Starred 2 days ago by

Eric Zhang(Founding Engineer at Modal). Also starred by

Lianmin Zheng(Coauthor of SGLang, vLLM),

Zhiqiang Xie(Coauthor of SGLang),

Ying Sheng(Coauthor of SGLang), and

1 more.

SpecForge by sgl-project

4.3%

964

Train speculative decoding models for faster inference

Supports online and offline training modes for flexibility.
Seamlessly integrates with SGLang for optimized serving.
Enables customization of training arguments, chat templates, and models.
Offers tensor parallelism and FSDP for large-scale training.

Created 1 year ago

Updated 9 hours ago

Starred 2 days ago by

Jeff Hammerbacher(Cofounder of Cloudera). Also starred by

Théophile Gervet(Cofounder of Genesis AI),

Jason Knight(Director AI Compilers at NVIDIA; Cofounder of OctoML),

Simon Mo(Core Maintainer of vLLM), and

6 more.

lingua by facebookresearch

0.0%

LLM research codebase for training and inference

Minimal PyTorch library; enables end-to-end training, evaluation.
Supports data parallelism, FSDP, model parallelism, activation checkpointing, float8.
Includes tooling for profiling MFU/HFU, memory usage, checkpoint management.
Provides example SLURM scripts for distributed training and evaluation.

Created 1 year ago

Updated 11 months ago

...