2024 Q3 Most Exciting LLM-powered Projects
Following up on the list of 2023’s, Q1 2024’s, Q2 2024’s most exciting projects, here is the list of the most useful, innovative and exciting LLM-powered projects I found in 2024 Q3.
GraphRAG: Boost Your LLM’s Performance with GraphRAG ❇️
Microsoft’s GraphRAG repository offers a cutting-edge, graph-based Retrieval-Augmented Generation (RAG) system designed to enhance the reasoning capabilities of large language models (LLMs) with structured data extracted from unstructured text1. By leveraging knowledge graphs, GraphRAG significantly improves the accuracy and depth of responses to complex queries, making it an invaluable tool for enterprises dealing with proprietary data2. Its modular design and user-friendly Solution Accelerator package on Azure ensure a seamless integration and deployment experience3. If you’re looking to boost your LLM’s performance in handling intricate datasets, GraphRAG is a must-try solution.
EDA GPT: LLM-powered data analysis 📊
EDA GPT is an innovative open-source data analysis tool that streamlines the process of exploring and gaining insights from both structured and unstructured data. What sets it apart is its ability to handle a wide range of data formats, including CSV, XLSX, SQLite, PDFs, and images, while leveraging advanced language models for intelligent analysis. The tool offers a user-friendly interface that allows users to interact with their data using natural language queries, generate visualizations, and obtain context-rich analysis reports. EDA GPT’s architecture incorporates sophisticated features such as multimodal search capabilities, advanced RAG techniques, and a unique “Lang Group Chain” for handling complex queries. With support for various LLMs, including offline options, and the ability to clean and edit data automatically, EDA GPT empowers users of all skill levels to make data-driven decisions efficiently and effectively. Demo video available here.
Composio: AI Agent framework with wide range of tools 🦾
Composio is a powerful toolset for AI agents that offers seamless integration with over 100 tools across various categories, including software platforms, operating systems, browsers, and more. Developers should consider using Composio for its ability to equip AI agents with high-quality tools and integrations without the hassle of managing authentication, accuracy, and reliability issues. A key selling point is its support for multiple agent frameworks like OpenAI, Claude, and Langchain, allowing developers to implement these tools with just a single line of code. Additionally, Composio now supports JavaScript, expanding its accessibility and making it an even more versatile choice for developers looking to enhance their AI agent capabilities across different programming environments.
LiteLLM: LLM APIs is OpenAI format 🌐
We all have screamed at some point because a new model provider does not follow the OpenAI SDK format for text, image, or embedding generation. Now you can use LiteLLM to call all LLM APIs using the OpenAI format. Use Bedrock, Azure, OpenAI, Cohere, Anthropic, Ollama, Sagemaker, HuggingFace, Replicate (100+ LLMs).
Agentscope: Easy LLM powered agents 🤖
AgentScope is an innovative platform designed to simplify the development of multi-agent applications powered by large language models (LLMs). It offers a user-friendly drag-and-drop interface, comprehensive documentation, and a variety of pre-built components, making it accessible even for developers with limited experience in multi-agent systems. With features like prompt tuning, streaming mode, and support for various OpenAI models, AgentScope enables the creation of sophisticated, scalable applications. This repository is ideal for a tech blog article as it showcases cutting-edge advancements in AI and provides practical tools for developers to experiment with and implement multi-agent solutions.
Llama recipes: Examples using Llama models 👩🏼🍳
Llama Recipes is a comprehensive repository designed to facilitate the fine-tuning of Meta’s Llama models, including the latest Llama 3.1. It provides a scalable library with scripts and notebooks for various use cases such as summarization, question answering, and domain adaptation. The repository supports both default and custom datasets and offers solutions for running Llama models locally, in the cloud, or on-premises. With its detailed examples and support for composable FSDP and PEFT methods, Llama Recipes is an excellent resource for developers looking to leverage advanced language models in their applications1. This makes it a valuable addition to any tech blog article focused on AI and machine learning advancements.
Gemma 2 2B: Compact & Powerful 💎
In a departure from our usual coverage of large language models (LLMs), we’re excited to highlight Google DeepMind’s Gemma 2 2B, a compact AI model that punches well above its weight class. This 2 billion parameter model offers a compelling blend of performance and efficiency that makes it accessible for a wide range of commercial PCs.
- Impressive Performance: Despite its relatively small size, Gemma 2 2B outperforms larger models like GPT-3.5, scoring 1130 on the Chatbot Arena benchmark.
- Efficiency at Scale: The model processes up to 2 trillion tokens, demonstrating robust capabilities across various applications.
- Hardware Versatility: Optimized for different hardware configurations, Gemma 2 2B is suitable for edge devices and laptops, making AI more accessible than ever.
- Open Source and Commercial-Friendly: Available under Gemma terms, the model can be freely used for both research and commercial applications.
- Ideal for On-Device AI: Gemma 2 2B excels in scenarios where hardware resources are limited, opening up new possibilities for local AI processing.
For developers and businesses looking to implement powerful AI capabilities without the need for extensive computing resources, Gemma 2 2B presents an exciting opportunity. Its ability to run efficiently on most commercial PCs makes it a game-changer for democratizing AI technology.
ClaudeDev: VSCode plugin autonomous engineer 🚀
Claude Dev is an open-source AI tool that acts as an autonomous software engineer within Visual Studio Code, making it a game-changer for developers. By automating complex tasks like reading and writing files, analyzing code, and executing terminal commands with user supervision, it significantly boosts productivity and ensures control and security. Its human-in-the-loop GUI allows developers to safely explore AI’s potential while maintaining oversight, making it an invaluable asset for enhancing coding efficiency and innovation. Source code.
Huggingface assistants: Unlocking Custom AI Power 🤗
Huggingface Assistants on HuggingChat are a dynamic and customizable way to create and share AI-powered tools with a community. They allow users to package prompts with different open-source models, offering flexibility in how they interact with content and provide answers. With recent updates, Assistants can now connect to the Internet, allowing for real-time data retrieval and more contextually relevant responses. This feature supports various modes, from general web searches to domain-specific queries and direct URL access, making it an ideal choice for users who want to tailor their AI’s knowledge base or functionality. Whether for casual use, specialized tasks, or interactive projects, Huggingface Assistants provide a powerful, open-source platform to enhance your AI experience and share it with others.
MindSearch: Autonomous (re)search engine🔎
MindSearch is an innovative, open-source AI search engine framework designed to mimic human-like search capabilities, making it an excellent choice for developers looking to build advanced search solutions. Leveraging large language models (LLMs) and a multi-agent architecture, MindSearch excels in decomposing complex queries into manageable sub-queries, ensuring comprehensive and accurate information retrieval. Its dynamic graph construction process and detailed solution paths enhance the depth and breadth of responses, outperforming other AI search engines like ChatGPT-Web and Perplexity.ai Pro. With support for both closed-source and open-source LLMs, and optimized user interfaces, MindSearch offers flexibility and superior performance, making it a compelling tool for creating sophisticated, user-centric search applications.
v0.dev: Chat based web development companion
v0.dev by Vercel is a game-changer for developers looking to streamline their UI creation process. This AI-powered tool generates ready-to-use React components using shadcn/ui and Tailwind CSS from simple text prompts. It’s perfect for quickly prototyping or enhancing your projects without getting bogged down in repetitive coding tasks. With v0.dev, you can focus more on innovation and less on the nitty-gritty details, making it a smart choice for efficient and effective web development. Checkout their gallery for inspiration or checkout this tutorial to find out how you can combine cursor with v0.
OpenHands: Open source Devin alternative
OpenHands, is an essential tool for anyone developing AI agents, boasting nearly 30k stars on GitHub. This platform enables the creation of generalist agents that can write code, fix bugs, and ship features, mimicking human developers. It offers a robust interaction mechanism for agents, interfaces, and environments, a sandboxed OS and web browser for safe code execution, and multi-agent support. With its comprehensive evaluation framework and the ability to collaborate with human developers, OpenHands is a powerful resource for advancing AI development.
MetaGPT: Python multi-agent framework
MetaGPT is an impressive Multi-Agent Framework with over 40K stars on GitHub. This innovative platform can simulate roles such as engineers, product managers, architects, and project managers. With just a single line of text, MetaGPT can generate the entire workflow of a software company, including data structures, APIs, documents, user stories, competitive analysis, and requirements. By integrating Standardized Operating Procedures (SOPs) into its processes, MetaGPT ensures efficient and structured workflows. It also supports incremental development, multilingual capabilities, and multiple programming languages, making it a versatile tool for collaborative AI development.
GPT Pilot: Your new agentic AI developer companion
The GPT Pilot repository by Pythagora-io represents a significant leap in AI-driven software development, offering a robust tool that goes beyond mere code generation to build fully functional applications. As the core technology behind the Pythagora VS Code extension, GPT Pilot serves as a comprehensive AI developer companion capable of writing, debugging, and discussing code with developers. This open-source project aims to demonstrate the potential of large language models (LLMs) in automating up to 95% of the coding process, leaving the remaining 5% for human oversight to ensure production-ready quality.
More updates coming soon
Revisit this article for more updates and let me know in the comments if you think I missed anything interesting.
