Projects

### Overview: This curriculum is designed to provide foundational knowledge in machine learning, focusing on language models. It aims to equip new Elicit employees with the necessary background to effectively deploy and scale ML in production. ### Key Features: - **Introduction to Machine Learning**: Covers the basics of machine learning, including neural networks and backpropagation. - **Transformers**: Provides a deep dive into large language models, including GPT and its components. - **Key Foundation Model Architectures**: Explains key architectures like Transformers, GPT, and others. - **Training and Finetuning**: Discusses techniques for training and fine-tuning models. - **Reasoning and Runtime Strategies**: Covers in-context reasoning, task decomposition, debate, tool use, and honesty in AI. - **Applications**: Explores how ML models are applied in science, forecasting, search ranking, and more. - **ML in Practice**: Discusses production deployment, benchmarks, datasets, and advanced topics like world models, planning, uncertainty, and interpretability. - **Advanced Topics**: Covers world models, causality, planning, uncertainty calibration, active learning, and reinforcement learning. - **The Big Picture**: Discusses AI scaling, AI safety, economic and social impacts, and philosophical considerations. ### Tech Stack: Python and Jupyter Notebooks. ### Why It Matters: This curriculum is crucial for anyone interested in the field of machine learning, particularly those working with language models. It provides a solid foundation in the principles and techniques used in modern ML, making it essential for anyone looking to advance in the field

Overview: Inkeep Agents is a powerful platform that allows teams to create and manage AI agents using either a no-code visual builder or a TypeScript SDK. The platform supports full 2-way sync and is compatible with Vercel and Docker for easy deployment. Key features: - No-Code Visual Builder: A drag-and-drop canvas for creating and managing Agents. - TypeScript Agents SDK: A code-first framework for building Agents with typesafety and intellisense. - 2-way sync: Technical and non-technical teams can edit Agents in either format. - Multi-agent architecture: Supports teams of agents for complex workflows. - MCP Tools: Credential management and dynamic chat experiences. - Observability: Traces UI and OpenTelemetry for easy debugging. - Managed and Enterprise options: Inkeep Cloud waitlist and Inkeep Enterprise. Tech stack: TypeScript, Vercel AI SDK, OpenTelemetry, Elastic License 2.0. Why it matters: Inkeep Agents provides a unified platform for creating and managing AI agents, making it accessible to both technical and non-technical teams. The platform's interoperability and 2-way sync feature allow for seamless collaboration and management across different teams and environments.

### Overview: This comprehensive guide serves as a step-by-step introduction to RAG (Retrieval-Augmented Generation) from basic concepts to advanced applications. It covers the fundamentals of RAG, including its working principles, benefits, and challenges, as well as detailed explanations of evaluation metrics and practical implementations using various libraries such as LangChain, Llama Index, and Haystack. ### Key Features: - **What is RAG?** - A simple example explaining RAG. - **Why RAG?** - Discusses the drawbacks of LLMs and how RAG addresses them. - **How does RAG work?** - Explains the indexing, retrieval, augmentation, and generation steps. - **RAG Benefits and Challenges** - Discusses the advantages and challenges of RAG. - **RAG Must Know Terms** - Definitions of RAG terms. - **RAG Roadmap** - A detailed plan to learn RAG from basics to advanced. - **RAG Developer's Stack** - Libraries used to build RAG systems. - **RAG from Scratch** - Implementation from scratch without any frameworks. - **RAG with LangChain** - Implementation using the LangChain framework. - **Website RAG** - RAG implementation over a website using LangChain. - **YouTube Video RAG** - RAG implementation over a YouTube video transcript using LangChain. - **Agentic RAG** - Agentic RAG system implemented using the CrewAI framework. ### Tech Stack: - **Jupyter Notebook** - Primary language used for the guide. ###

Overview: Korvus is a search SDK that unifies the entire RAG pipeline into a single database query, built on top of Postgres with bindings for Python, JavaScript, Rust, and C. It simplifies the search architecture by providing high-performance, customizable search capabilities with minimal infrastructure concerns. Key features: - **Simplified Architecture**: Replace complex service-oriented architectures with a single, powerful query. - **High Performance**: Eliminates API calls and data movement for faster processing and greater reliability. - **Open Source**: Improve your developer experience with open source software and models that run locally in Docker. - **Multi-Language Support**: Use Korvus with Python, JavaScript, and Rust. Open an issue to vote for other language support. - **Unified Pipeline**: Combine embedding generation, vector search, reranking, and text generation in one query. - **Postgres-Powered**: Under the hood, Korvus operations are powered by efficient SQL queries on a time-tested database platform. Tech stack: Rust Why it matters: Korvus simplifies the search architecture by unifying the entire RAG pipeline into a single database query, leveraging Postgres' robust capabilities for efficient and reliable performance.

Overview: CLI Agent Orchestrator (CAO) is a tool designed to coordinate multiple AI coding CLIs, enabling a supervisor to delegate tasks to specialized agents in parallel or sequence. Key features: - Starts multiple AI coding CLIs in isolated terminal sessions - Provides a supervisor with tools for coordinating workers - Runs as full CLI processes with native authentication and capabilities - Supports multiple provider CLIs, including Claude Code, Kiro, Codex, and more Tech stack: Python Why it matters: CAO simplifies the coordination of AI coding tasks, allowing for efficient parallel or sequential processing of work, making it particularly useful in environments where multiple specialized agents are needed. Repository: awslabs/cli-agent-orchestrator Primary language: Python GitHub stars: 1460 GitHub description: Multi-agent orchestration for AI coding CLIs — Claude Code, Kiro, Codex, and more, coordinated in isolated tmux sessions

### Overview: The LabelMe annotation tool source code is a JavaScript-based application designed for online image labeling. It provides a convenient way for users to annotate images without the need to install or copy a large dataset onto their computers. ### Key Features: - **LabelMe Annotation Tool**: The primary feature of the tool, allowing users to annotate images online. - **Image Storage**: Images are stored in the "Images" folder. - **Annotations**: Annotations are collected in the "Annotations" folder. - **Masks**: Segmentation masks are stored in the "Masks" folder. - **Scribbles**: Scribbles are collected in the "Scribbles" folder. - **Tool Page**: The main web page for the LabelMe annotation tool is located in the "tool.html" file. - **Tool Directory**: Source code is located in the "annotationTools" directory. - **Temporary Files**: Temporary files are stored in the "annotationCache" directory. - **Icons**: Icons used on the web page are located in the "Icons" directory. - **Docker Files**: Instructions for Docker installation are provided in the "DockerFiles" directory. ### Tech Stack: JavaScript ### Why it Matters: The LabelMe annotation tool offers a more accessible and flexible way to annotate images compared to traditional image annotation tools. It allows users to access the tool from anywhere, making it ideal for collaborative annotation projects.

### Overview: Nixopus is a first-of-its-kind open-source deployment platform that automates the entire deployment lifecycle, from analyzing code to deploying apps. It uses an autonomous AI agent to handle everything, from generating configurations to fixing issues, making it fully automated and efficient. ### Key Features: - **AI-Powered Lifecycle**: The agent generates configs, deploys, and fixes failures autonomously, closing the loop on failed deployments. - **Chat Interface**: A conversational interface for deploying, adding domains, checking logs, rolling back, and troubleshooting. - **Editor Extension**: Integrates with popular editors to deploy code changes directly. - **Multi-Server Orchestration**: Supports seamless deployment across multiple servers. - **Full Machine Access**: Provides full machine access for developers to interact with the platform. - **Framework Agnostic**: Works with any framework, ensuring compatibility with various applications. ### Why It Matters: Nixopus simplifies the deployment process by automating everything, reducing the need for manual intervention and minimizing errors. It's particularly useful for developers and teams looking to streamline their deployment workflows and focus on their core responsibilities. ### Tech Stack: Go ### GitHub Stars: 1470 ### GitHub Description: Run production apps without thinking about infrastructure. On your server or ours. Fully agentic. ### Demo: See Nixopus in action, from deploying apps to day-to-day operations on your infrastructure. ### Getting Started & Join Our Community: - **How It Works**: Connect your repo, tell the agent to deploy, and go live. - **Join Our Community

**Overview:** Agor is a multiplayer AI platform designed to bring your team and agents together on live spatial boards. It allows you to see the work, collaborate, and learn from each other, making AI collaboration a team activity. **Key features:** - Live cursors, comments, and shared sessions for colleagues to follow an agent's progress. - See how colleagues prompt, share context, and turn practices into reusable prompts and workflows. - Start solo and expand as your team grows. **Tech stack:** TypeScript. **Why it matters:** Agor simplifies AI collaboration by bringing the work of multiple agents into a single, shared workspace, making it easier for teams to learn from each other and build upon each other's progress. Repository: preset-io/agor Primary language: TypeScript GitHub stars: 1464 GitHub description: Multiplayer AI: Work together again.

Overview: ClawTeam-OpenClaw is a fork of the ClawTeam project fully adapted for OpenClaw, featuring multi-agent swarm coordination using OpenClaw as the default agent. Key features: - Multi-agent swarm coordination - OpenClaw as the default agent - Per-agent session isolation - Exec approval auto-config - Production-hardened spawn backends Tech stack: Python Why it matters: ClawTeam-OpenClaw simplifies multi-agent swarm coordination by integrating OpenClaw as the default agent, providing a robust and production-hardened solution for agents to self-organize and collaborate effectively. Repository: win4r/ClawTeam-OpenClaw GitHub stars: 1467

### Overview: LanPaint is a universal inpainting sampler that supports various tasks such as training-free image inpainting, outpainting, and local editing. It leverages pretrained diffusion and rectified-flow models to achieve superior quality without fine-tuning, making it highly versatile and powerful. ### Key features: - **Training-Free Inpainting**: LanPaint enables image inpainting without the need for fine-tuning or backpropagation. - **Outpainting and Generative Fill**: Supports outpainting and generative fill, allowing for more complex image editing. - **Mask-Constrained Local Editing**: Enables object/region replacement with text guidance and character-consistent local generation. - **Video Inpainting and Local Video Editing**: Supports video inpainting and local video editing. - **Video + Audio Masked Generation**: Enables the generation of videos with masked audio. ### Tech stack: Python ### Why it matters: LanPaint is a powerful tool for image editing and inpainting tasks, making it highly versatile and efficient. Its training-free approach and ability to handle complex tasks make it a valuable addition to any developer's toolkit.

Overview: `aeon` is a Python library designed for time series machine learning and deep learning, built by the researchers behind the algorithms. Key features: - Classification - Regression - Clustering - Forecasting - Anomaly Detection - Distance Metrics - Segmentation - Similarity Search - Transformations - Benchmarking Tech stack: Python Why it matters: `aeon` provides researchers and practitioners with a comprehensive toolkit for time series analysis, featuring state-of-the-art models and contributions from the creators of the original methods.

### Overview: This repository is a curated list of recent Transformer-based works in computer vision, focusing on recent papers published in the field. It serves as a survey of advancements and applications of Transformers in various CV tasks. ### Key features: - **Action**: Techniques for action recognition and generation. - **Active Learning**: Methods for improving model performance through active learning. - **Adversarial Attacks**: Strategies to defend against adversarial attacks in computer vision. - **Anomaly Detection**: Algorithms for detecting anomalies in images and videos. - **Assessment**: Metrics and evaluation methods for assessing model performance. - **Augmentation**: Techniques for enhancing image quality and diversity. - **Audio**: Applications of Transformers in audio processing. - **Bird's-Eye-View**: Methods for generating bird's-eye-view images. - **Captioning**: Transformers for image captioning tasks. - **Change Detection**: Techniques for detecting changes in images. - **Classification (Backbone)**: Classification models using Transformers as backbones. - **Clustering**: Methods for clustering images and data. - **Completion**: Techniques for filling in missing parts of images. - **Compression**: Methods for reducing the size of image representations. - **Cross-view**: Models that can handle multiple view inputs. - **Crowd**: Techniques for crowd counting and tracking. - **Deblurring**: Methods for removing blurriness from images. - **Depth**: Transformers for estimating depth from images. - **Deepfake Detection**: Algorithms for detecting deepfake videos. - **Dehazing**: Techniques for removing haze from images. - **Deraining

Overview: Jonex is an all-in-one multimodal parsing engine and AI-ready knowledge engine designed for enterprise agents. It integrates data ingestion, multimodal parsing, domain knowledge compilation, vector and graph indexing, source-grounded retrieval, and feedback loops to create reusable knowledge services. Key features: - Multimodal parsing engine - AI-powered knowledge engine - Domain ontology integration - Vector and graph indexing - Source-grounded retrieval - Multi-tenant ready Tech stack: Python, React, FastAPI, Docker Compose, Multi-tenant ready Why it matters: Jonex simplifies the creation and management of enterprise knowledge by turning raw content into actionable knowledge, making it easier for enterprise agents to access and utilize the knowledge they need. GitHub stars: 1445

### Overview: Irodori-TTS is a high-performance Text-to-Speech (TTS) model designed for generating natural-sounding voices from text inputs. It leverages Flow Matching and continuous DACVAE latents to create a more lifelike voice output. The model supports various features such as voice cloning, multi-modal voice design, and automatic duration prediction, making it versatile for both training and inference. ### Key Features: - **Flow Matching TTS**: Rectified Flow Diffusion Transformer (RF-DiT) over continuous DACVAE latents. - **Voice Cloning**: Zero-shot voice cloning from reference audio. - **Multi-modal Voice Design**: Combines text, reference speech, and caption text for voice identity and style control. - **Long Reference Audio**: Supports up to 120 seconds of reference audio. - **Emoji-based Style Control**: Influences the delivery and non-verbal vocal expressions through emoji annotations. - **Automatic Duration Prediction**: Estimates output length without manual `--seconds`. - **Automatic Watermarking**: Generated audio is watermarked with SilentCipher when available. - **Multi-GPU Training**: Distributed training with gradient accumulation, mixed precision (bf16), and logging with W&B. - **PEFT LoRA Fine-Tuning**: Parameter-efficient adaptation with PEFT/LoRA for released checkpoints. - **Speaker Inversion**: Freezes the base model while learning reusable speaker embedding tokens. - **Flexible Inference**: CLI, Gradio Web UI, and HuggingFace Hub checkpoint support. ### Tech Stack: The model is

### Overview: This repository contains the code for a baseline model designed for 3D human pose estimation. The model was presented at the International Conference on Computer Vision (ICCV) in 2017 and is based on the paper "A simple yet effective baseline for 3D human pose estimation." The project is described in detail in the accompanying paper and video. ### Key features: - The repository includes the code for the ICCV 2017 paper. - Dependencies for running the code include Python 3.5 or later, cdflib, and TensorFlow 1.0 or later. - The data for training and testing is provided, including the H3.6M dataset. - A quick demo script is included to train the model for one epoch and visualize the results. - The code is designed to be transparent, compact, and easy to understand. ### Tech stack: Python 3.5 or later, cdflib, TensorFlow 1.0 or later ### Why it matters: The model is lightweight and provides a strong baseline for 3D human pose estimation, highlighting the challenges of current approaches. The code is transparent, compact, and easy to understand, making it suitable for educational and research purposes.

Overview: This repository is designed to provide a comprehensive, hands-on guide for learning AI from scratch, covering math fundamentals, PyTorch, deep learning, transformers, retrieval-Augmented Generation (RAG), optical character recognition (OCR), and additional modules such as LLM, fine-tuning, basic ML model implementation, DL projects, and data analysis. Key features: - Math functions, derivatives, vectors, and gradients - Matrix operations and linear algebra - Probability and statistics - PyTorch basics - Neural-Network(NN) - Transformers - Retrieval-Augmented Generation (RAG) - Optical Character Recognition (OCR) - Additional modules including LLM, fine-tuning, basic ML model implementation, DL projects, and data analysis Tech stack: Jupyter Notebook Why it matters: This repository serves as a complete, hands-on guide to learning AI from first principles, building real neural networks, and understanding modern LLM systems end-to-end. It provides a clear, structured, and intuitive approach to building real AI systems, making it suitable for both beginners and experienced engineers. Repository: Ramakm/AI-Engineering GitHub stars: 1462 GitHub description: A group of notebooks and other files which can help you learn AI from scratch.

### Overview: Knowledge is a powerful tool for saving, searching, accessing, exploring, and chatting with your favorite websites, documents, and files. ### Key Features: - **Knowledge Graph**: Visualize connections between your data. - **Built-in Chromium**: Access built-in Chromium for summarizing, extracting topics, and turning web content into notes. - **Chat Feature**: Engage in dynamic conversations with Projects and Sources, leveraging Large Language Models. ### Tech Stack: TypeScript ### Why it Matters: Knowledge transforms how you interact with your data, offering a more engaging and exploratory approach to learning. Whether you're seeking to better understand complex topics or exploring new ideas, the chat feature is your personal knowledge companion.

Overview: MLReef is a collaborative ML-Ops platform designed for sharing and managing Machine Learning work. Key features: - Data Management: Fully versioned data hosting and processing infrastructure. - Publishing Code Repositories: Containerized and versioned script repositories for immutable use. - Experiment Manager: Experiment tracking, environments, and results. - ML-Ops: Pipelines & Orchestration solution for ML/DL jobs (K8s / Cloud / bare-metal). Tech stack: Kotlin. Why it matters: MLReef streamlines the Machine Learning Development Lifecycle, making collaboration and reproducibility easier and more efficient.

Overview: GaussianEditor is a cutting-edge 3D editing tool that enables swift and controllable manipulation of images and videos, leveraging Gaussian splatting for precise and efficient editing. Key features: - **Gaussian Splatting**: A powerful technique for editing images and videos by applying Gaussian blur to specific areas. - **Swift Editing**: Achieves real-time results with minimal user interaction. - **Controllable Editing**: Allows users to precisely control the editing process with adjustable parameters. Tech stack: C++ Why it matters: GaussianEditor offers a highly efficient and intuitive solution for 3D editing, making it a valuable tool for both professionals and enthusiasts in the field of computer vision and image processing. Repository: buaacyw/GaussianEditor Primary language: C++ GitHub stars: 1462 GitHub description: [CVPR 2024] GaussianEditor: Swift and Controllable 3D Editing with Gaussian Splatting

Overview: Alook is a platform where people and AI agents can collaborate in shared rooms, allowing users to reach their agents directly and keep them running on their machines. Key features: - Collaboration: Invite trusted people into a room where agents can talk with them. - Local-first & Always-on: Agents have persistent identities and stay on the user's machine. - One identity: Agents have a unique identity across every room. - Memory with initiative: Agents move work forward and keep tasks organized. - Reach: The same Alook room updates across desktop and mobile. Tech stack: TypeScript Why it matters: Alook simplifies collaboration between people and AI agents, making it easier for teams to work together seamlessly and keep their agents running on their machines.