Inkling Labs

1) Motivation We're building Inkling, an AI-enabled data studio where ANYONE can organize, clean, annotate, and prepare high-quality datasets for frontier labs, without needing expertise in data engineering. Today, turning this data into high-quality AI training datasets often requires specialized expertise or partnerships with companies like Scale AI or Mercor. These services can be expensive and difficult for smaller organizations to access, which leaves smaller research institutions and independent experts largely excluded from the AI data economy. We want to democratize data labelling by giving anyone the tools to prepare and annotate their data without needing a dedicated data engineering or machine learning team. We are starting with medical imaging, where hospitals, clinics, and researchers work with large amounts of valuable, complex data but often lack accessible tools to prepare it for AI. Our long-term vision is to make working with data as simple as working with documents, and that no one should be excluded from the data economy. 2) Features - Upload different types of data, including X-rays, CT scans, MRIs, ECGs, pathology slides, videos, text, and sensor data. Files are checked for PII information before being made available in the Studio. - Describe what you want to do in plain English, and our agent automatically creates the annotation environment. - Our plugin-based architecture supports different types of data, each with its own specialized viewer and annotation tools. This also makes it easy to add support for new data types. - Meta SAM 3.1 helps speed up medical image annotation by automatically suggesting segmentation masks that users can review, edit, or accept. - Annotations are version-controlled, with branching and review workflows so teams can collaborate, track changes, and see who contributed each label. - Privacy is built into the platform. The AI assistant only sees file metadata, not the actual contents. Data is stored in a separate cloud project, and AI-generated changes require user approval before they are applied. 3) Tech Stack - Frontend: React 19, Vite, TypeScript, Cornerstone3D, vtk.js - AI Models: Meta Muse Spark 1.3 (AI assistant), Meta SAM 3.1 (image segmentation), Meta Model API - Agent Infrastructure: Node.js 24, pi agent harness, Model Context Protocol (MCP), Daytona sandboxes - Backend: Python, FastAPI, Python workers, Go (dataset versioning) - Database & Storage: PostgreSQL 17, Google Cloud Storage, Redis - Cloud & Infrastructure: Google Cloud Platform (GCP), Google Kubernetes Engine (GKE) Autopilot, Terraform - Authentication: WorkOS 4) Try Inkling Website: https://inklinglabs.io/ To try Inkling, please sign in through WorkOS. We use WorkOS to manage user authentication and sessions, and to help prevent abuse of the platform. Once in, you can access the Studio and create your own annotation environment.