AI Makerspace

🚩 AI Makerspace is on a mission to create the world's leading community for people who want to build, ship, and share production LLM applications. We are committed to open-sourcing 📚 practical AI education that 🍋 makes it easy for people to start 🏗️ building, 🚢 shipping, and 🚀 sharing, and to 🛣️ embrace the LLM edge. In that spirit, we have decided to 🫴 give first and open-source all of the concepts and code from LLM Engineering - Foundations to SLMs for the entire community. This was a $1999/seat value during the holidays last year! 🧑💻 In addition to concepts and slides, you get access to all project assignment builds. S/o to all the folks who participated in Cohort 3 of LLM Engineering - Foundations to SLMs last November-December of 2024! Without you this open-sourcing wouldn't be possible 🙏. Start learning, building, shipping, and sharing now! - Everything in one place: https://lnkd.in/gr-txmQ9 - YouTube: https://lnkd.in/gTy5dC2j - GitHub: https://lnkd.in/gURQxEQk PS... special s/o to our team behind the scenes who helped bring this production and open-source project together, especially Sarah deSouza, Chris Lusk🕵️♂️, and Jacob Kilpatrick as well as all of the peer supporters Rodrigo Díaz Concha Debora Andrade, PhD Nithin Kamavaram and Allan Tan for providing amazing mentorships to our students along the way! #llmengineering

We are thrilled to announce that Cohort 5 of The AI Engineering Bootcamp is gearing up for an epic Demo Day! 🏗️ 🚢 🚀

What we’re 🏗️ building, 🚢 shipping, and 🚀 sharing this week: Optimization of LLMs: Series Introduction 📈This week, we kick off a brand new series on what it means to optimize the performance of an LLM. 🔄 PPO, DPO, GRPO… we hear a lot about how to “optimize” transformers today, whether we’re aligning them, tuning them for reasoning, or we have some other optimization objective. 🎮 Setting up architectures and then optimizing their performance has always been the name of the game in Machine Learning and Deep Learning. 📉Going from noisy, untrained models to AI systems that can predict, generate, and even reason has been a long road. How did we get here? This new series will connect the dots—from classical optimization to the latest breakthroughs in training LLMs. If you’ve ever wondered why models are trained the way they are, this is for you. 📚 You’ll learn: 🗺️ The evolution of optimization techniques prior to ML 🤖 How we got from gradient descent to transformers 🔄 How we can understand modern optimization schemes Join us this Wednesday at 10 AM PT to kick off!  🔗 https://lnkd.in/gaK4Rz5g #AIEngineering #MachineLearning #DeepLearning #Optimization #LLMs

What we built 🏗️, shipped 🚢, and shared 🚀 last week: PydanticAI! Q: "If my team uses Pydantic, should we use PydanticAI for agents?" A: Maybe Consider these additional questions: 1️⃣Does your team enjoy “that FastAPI feeling?” 2️⃣Do you mind having an agent code base that, compared to other, has quite a bit of code (albeit Pydantic-style)? 3️⃣Are you a fan of amazing and hilarious docs? In the end, PydanticAI: ✅ Does the thing! We can build agents and mulit-agent systems 🛣️ It’s still early in their journey (beta version, feedback welcome) 🏄 Great open-source vibes, committed to OS, and great docs Do you think your team can move faster by picking up this framework over another one?  Let us know why or why not in the chat 👇 🎦 https://lnkd.in/eK4Es2UA #PydanticAI #Agents #Orchestration

Pydantic AI has jokes! Follow along as The Wiz demos the how to build a multi-agent joke app with Pydantic! #learnai #pydantic #genai Watch the full live stream at https://lnkd.in/effcfDAZ

“I can’t build right now because I’m looking for a job full-time.” Please don’t do this. Just 🏗️ build, 🚢 ship, and  🚀share, everyone. It’s really not that hard or complicated.

Building a RAG Pipeline with Hugging Face Endpoints and LangChain I am thrilled to announce that I have just built and shipped a open-source LLM-powered RAG Application with LangChain! 🎉🤖 Why to use open-source LLM and embedding model endpoints? This would allow customers to deal with private data in a secure and scalable way!!! The open-source LLM models are hosted on Hugging Face's inference endpoints. This involved leveraging LangChain components, like HuggingFaceEndpoint and HuggingFaceInferenceAPIEmbeddings, to seamlessly integrate with the endpoints. The process included: 1. Choosing correct open source LLM and embedding model. 2. Setting up the hugging face inference endpoint with above models. 3. Building a prototype to use the deployed models 4. Converting the prototype to a chainlit application and making it available to public through HuggingFace space. GitHub link for prototype: https://lnkd.in/gRjMkt-q Loom video explaining the work: https://lnkd.in/gAZHhMVB Hugging face space link: https://lnkd.in/g_SY7Btt This hands-on experience provided valuable insights into building efficient and scalable RAG pipelines using open-source tools. The combination of Hugging Face's inference endpoints and LangChain's flexibility opens up exciting possibilities for leveraging powerful language models in various applications. 📚 Lessons Learned: 1. Leveraging Hugging Face Endpoints for RAG: How to use Hugging Face Inference Endpoints for both LLMs and Embeddings within a RAG pipeline. This allows us to access and utilize powerful open-source models without needing to host them ourselves. 2. Use of open-source endpoints for production: How to choose the correct open-source model. 3. How to use the Llama 3.1 8b Instruct model: Adapting prompts based on how Llama Instruct was fine-tuned/trained. 🤔 Lessons Yet to Learn: 1. How to use other settings of HF inference endpoints, for example, tasks other than text generation. 2. How to configure monitoring on HF inference endpoints so that we get alerts whenever something is not working. I'm thrilled to continue this journey and explore more about production and open-source endpoints. If you have any insights or suggestions, feel free to share! Let's continue pushing the boundaries of what's possible in the world of AI and question-answering. Here's to many more innovations! 🚀 Shout out to AI Makerspace ! #LangChain #QuestionAnswering #RetrievalAugmented #Innovation #AI #TechMilestone Feel free to reach out if you're curious or would like to collaborate on similar projects! 🤝🔥

Excited to learn about Open Deep Research from AI Makerspace What is Open Deep Research? It’s a powerful multi-agentic framework that generates comprehensive reports on any topic with a workflow similar to OpenAI and Gemini Deep Research but with full customization. You can: • Provide an outline with your desired report structure • Set the planner model (e.g., DeepSeek, OpenAI reasoning model, etc.) • Give feedback on report sections and iterate until you’re satisfied • Choose your search API (e.g., Tavily, Perplexity) & number of searches per iteration • Control the depth of search per section (writing, reflection, search, re‑write cycles) • Customize the writer model (e.g., Anthropic) Watch the video to see how to harness these advanced features to transform your research workflows! #OpenDeepResearch #DeepResearch #AI #LLM #Innovation #Research #Tech #WebResearch #DeepLearning #Customization Alex Meehan Shahul ES

🚀 DeepSeek’s Open-Source Week: A Game Changer for LLMs! DeepSeek just dropped five powerful open-source repos—and we’re diving in! From FlashMLA for optimized decoding to Fire-Flyer File System (3FS) for blazing-fast data access, these releases push AI efficiency to the next level. Join us LIVE as we break down what’s inside, how it powers DeepSeek-R1, and what it means for the future of LLMs. No ivory towers—just pure innovation, open-source, and community-driven progress. 🔥 📅 Don’t miss it! Bring your questions and let’s explore together. Event Resources: 🙋♀️ Drop your questions during the event in Slido: https://lnkd.in/gh2Vd4Us 🤗 How'd we do? Share your feedback and suggestions for future: https://lnkd.in/gsB-vnDD 🧑🤝🧑 Join our community to start building, shipping, and sharing with us today: https://lnkd.in/gpf4xTyk 🗓️ Stay up to date with AIM Events: https://lu.ma/aimakerspace #DeepSeek #OpenSource #LLM

🚀 From Gradient Descent to GRPO: A Deep Dive into Optimization for LLMs We’ve all heard of RLHF, PPO, and DPO. And now, GRPO. We hear a lot about test-time compute vs. train-time compute. Historically, we’ve heard a lot about in-context learning. Even meta-learning. 🧵The common thread? Optimization. Every major breakthrough in the saga of LLMs, from reinforcement learning to transformers, has been about optimizing LLMs to perform better - from either a cost or performance perspective - at specific tasks. 🛣️ But how did we get here? 🔍 In our new series, we’ll trace the path from classic optimization to modern LLMs, covering: ⛰️The evolution from gradient descent to policy optimization (PPO, DPO, CGPO, GRPO) 🏁How deep learning milestones like CNNs, RNNs, and LSTMs paved the way for transformers ✅The journey from seq2seq to attention to reasoning This isn’t just theory—we’ll build, ship, and share real-world applications. Spoiler alert: we’re building black boxes and everything is loss. 🔗 Join us live to be part of the ongoing conversation! https://lnkd.in/gaK4Rz5g Bring your questions, and let’s break down these optimization techniques together. #LLMs #Optimization #DeepLearning #AI #MachineLearning