AI Repo: Shubhamsaboo/awesome-llm-apps

Skyler: Hey everyone, welcome back to Git News A I, the show where we dig through the hottest trending A I repositories on GitHub so you don't have to. Big thanks to our sponsors today — Ada, who helps teams integrate A I into their workflows seamlessly, Ago Consulting, your silicon development partners from A I all the way to SoC, and Zen Semiconductor, an umbrella company with a whole portfolio of silicon-industry companies, plus Ada for deploying A I at scale in any business, including their Vect appliance that runs local L L Ms and vector search on dedicated hardware. I'm Skyler, and today we've got five really solid repos to get through — we're talking massive L L M app collections, NVIDIA's transformer training infrastructure, A I security red teaming, official Claude plugins, and Mojo's Modular platform. Let's get into it. Skyler: Alright, kicking things off with a repo that honestly just keeps growing and growing — it's Shubhamsaboo's awesome-llm-apps on GitHub, and this thing has absolutely exploded. We're talking over 134,000 stars, which is a staggering number for an open source collection like this. So what is it? It's essentially a curated, free, and open source library of over a hundred A I agents, agent skills, and R A G applications. R A G, if you're new to the term, stands for Retrieval Augmented Generation — it's the technique where you connect an L L M to external data sources so it can answer questions based on your own documents or knowledge bases, rather than just its training data. And this repo doesn't just explain concepts — it gives you working implementations you can actually run. The breadth here is genuinely impressive. You've got agents for research, coding, customer support, finance, productivity — basically every domain you can think of. Each app is self-contained, documented, and designed to be a learning resource as much as a starting point for your own projects. If you're an M L engineer or developer trying to get your hands dirty with agentic A I, this is one of the best sandboxes available. The recent activity shows the maintainers are actively integrating sponsor and community contributions, which means the collection keeps expanding. Whether you want to understand how tool-calling agents work, how to build multi-step reasoning pipelines, or just want a reference implementation for a specific use case, this repo has something for you. Beginners will appreciate how approachable it is, and experienced practitioners will find patterns worth stealing for production projects. Seriously, if you haven't starred this one yet, just go do it — check the show notes for the link. Skyler: Next up, let's talk about something on the other end of the spectrum in terms of scale — NVIDIA's Megatron-L M. This one sits at around 17,500 stars, and while that might seem modest compared to our last repo, the impact of this codebase on the A I industry is absolutely enormous. Megatron-L M is NVIDIA's research framework for training transformer models at massive scale. We're talking the kind of scale that powers frontier L L Ms — billions, sometimes hundreds of billions of parameters. The framework has been instrumental in the development of some of the most capable language models in existence. So what makes it special? Megatron-L M implements several parallelism strategies that are critical for training at this scale. You've got tensor parallelism, which splits individual model layers across multiple G P Us, pipeline parallelism, which divides the model into stages across machines, and sequence parallelism for handling very long contexts efficiently. Combining all three is sometimes called 3D parallelism, and it's essentially the secret sauce that lets researchers train models that wouldn't even fit in memory on a single machine. The recent C I updates — bumping Cerno commits and adjusting Docker build provenance settings — tell you this is a living, actively maintained production-grade system, not just a research artifact. If you're working at a company or research institution doing serious pre-training or fine-tuning of large models, this is required reading. Even if you're not operating at that scale yet, understanding how Megatron-L M solves distributed training problems will make you a better M L engineer full stop. This is the kind of infrastructure knowledge that separates folks who train models from folks who really understand training. Check the show notes for the link. Skyler: Okay, shifting gears to something that's increasingly important as A I gets deployed everywhere — security. Tencent has open-sourced something called A I Infra Guard, and it just crossed nearly 6,000 stars, which is pretty impressive for a security-focused tool in this space. The description calls it a full-stack A I red teaming platform, and that framing is really apt. Red teaming, for the uninitiated, is basically ethical adversarial testing — you try to break your own systems the way an attacker would, before the attacker does. A I Infra Guard takes this concept and applies it specifically to A I ecosystems. It covers a lot of ground. There's an Agent Scan component that looks at how your A I agents behave under adversarial conditions. Skills Scan examines the individual capabilities or tools an agent can call. There's an M C P scan — M C P being the Model Context Protocol, a relatively new standard for how agents interact with tools and data sources. Then you've got A I Infra scan, which looks at your broader infrastructure, and L L M jailbreak evaluation, which tests whether your models can be manipulated into producing harmful or unintended outputs. The recent activity shows fixes around C V E count tracking and documentation cleanup, which tells you the team is actively maintaining a vulnerability database as part of this. This is exactly the kind of tooling that enterprise teams deploying A I need to take seriously. Regulatory pressure around A I safety is only going up, and having a systematic way to evaluate your attack surface is going to become table stakes. Security engineers, A I platform teams, and anyone responsible for deploying agents in production should absolutely have a look. Check the show notes for the link. Skyler: Alright, this next one is going to be interesting for anyone building on top of Anthropic's Claude — it's the official Claude plugins directory, anthropics/claude-plugins-official, sitting at over 34,000 stars. So this is Anthropic's own, managed directory of high-quality plugins specifically designed for Claude Code, which is their agentic coding product. Think of it like a curated app store for extending what Claude can do when it's acting as your coding assistant. The plugins in this directory are vetted and maintained by Anthropic themselves, which is a meaningful distinction — you're not just grabbing random community packages and hoping they work. The recent activity gives you a sense of what's being added: there are new NetSuite integrations, which is significant because NetSuite is a major enterprise ERP system, meaning Claude Code is getting real enterprise software capabilities. There's a Unity plugin for game development workflows, and a Qodo entry which is a code quality tool. This is telling us something important about where agentic A I is heading — it's not just about chat interfaces anymore, it's about deeply integrating with the specific tools and platforms developers actually use every day. If you're building with Claude Code or thinking about adopting it for your team, this directory is your starting point for understanding what's possible out of the box. And if you're a tooling vendor or developer, getting your integration listed here is probably a smart move from a distribution standpoint. The fact that Anthropic is officially curating this rather than leaving it fully to the community suggests they're serious about quality and reliability in the plugin ecosystem. Check the show notes for the link. Skyler: And last but definitely not least, let's talk about the Modular platform — modular/modular on GitHub — which includes both the MAX inference engine and the Mojo programming language. This one's sitting at nearly 30,000 stars and it's written primarily in Mojo, which is itself fascinating. For context, Mojo is a relatively new language designed by Modular to be a superset of Python that can achieve systems-level performance — we're talking C++ or even close to hardware-level speeds while still feeling like Python to write. The idea is to close the gap between the research world, which lives in Python, and the production world, which often needs performance that Python just can't deliver. MAX is their inference engine built on top of this foundation, designed to run models fast across different hardware targets. The recent activity shows version pinning for Mojo 1.1.0 dev builds and a freeze on sampling and model configs for MAX 26.6, which indicates they're on an active development cadence and moving toward stabilizing their configuration interfaces. This is the kind of platform-level work that has long-term implications for how A I models get deployed. If the Mojo language matures and the ecosystem grows, it could meaningfully shift how M L engineers think about the performance versus productivity tradeoff. For language nerds, systems programmers curious about A I, or anyone thinking seriously about inference optimization, this is a repo worth watching closely even if you're not ready to adopt it today. The trajectory is compelling. Check the show notes for the link. Skyler: And that's a wrap on today's repos — what a lineup. Huge thanks to our sponsors: Ada for helping teams actually integrate A I into real workflows, Ago Consulting for being the go-to experts in silicon development from A I to SoC, and Zen Semiconductor for pushing the hardware frontier with their portfolio of silicon companies and the Vect appliance bringing local L L Ms and vector search to dedicated hardware. If you want more A I and tech coverage, check out AI Special Report, another great show in the Ada podcast network that goes deep on the trends shaping the industry. And of course, subscribe to Git News A I wherever you get your podcasts — the show notes have all the links to every repo we talked about today, so you can jump straight in. I'm Skyler, and I'll catch you in the next one.