Ocular AI (YC W24)

We’re releasing our first open-source dataset: a 10-hour, 7,377-recording multi-accent English ASR dataset spanning 11 countries (China, Vietnam, Thailand, Japan, Russia, Poland, Eastern Europe, Indonesia, France, Germany, South Korea). Built on Ocular AI’s data infrastructure and global expert network, the dataset reflects how people actually speak—across accents, pronunciation, cadence, and linguistic nuance. It’s a step toward improving ASR performance across diverse accents and enabling more robust, human-like voice systems. Blog: https://lnkd.in/grTcxFq3 Dataset: https://lnkd.in/grMdEARi License: Open Data Commons Attribution License (ODC-By v1.0)

The Admin Console is now live on Foundry. Manage all your organization’s Ocular AI workflows from one centralized hub—complete visibility, full control. ✅ Monitor all workspaces and projects ✅ Track usage and performance ✅ Control access with Role-Based Access Control (RBAC) ✅ Manage API keys securely Built for enterprise teams that need governance at scale—without the overhead. Changeling: https://lnkd.in/gNRkDWnK

We just shipped a Quick Action dropdown on the workspace sidebar in Foundry. Access the most common workflows in Foundry instantly from the sidebar using the new Quick Actions dropdown. Changelog: https://lnkd.in/gNRkDWnK

Launch Week III, Day 5 — GPU Access & Model Training on the Multimodal Lakehouse 🚀 We’re excited to ship GPU Model Training on Ocular Foundry — enabling teams to go from curated multimodal data to fine-tuned AI models. No data movement. No complex setup. Just powerful, enterprise-grade GPU infrastructure directly integrated into your Multimodal Lakehouse. With this release, Ocular Foundry becomes a full-stack platform for building multimodal AI — from data ingestion, annotation, and analytics to training and deployment. 🔧 Train models like YOLO11 directly on versioned datasets ⚙️ Adjust hyperparameters and monitor training in real time 📈 View full training metrics — mAP, loss curves, confusion matrices 📁 Export model weights instantly, ready for deployment 🎥 Watch the demo: https://lnkd.in/gzRNU7Nv 📖 Read the blog: https://lnkd.in/gHfc-Wab 📅 Book a demo: https://lnkd.in/gKSA4UpT

Launch Week III, Day 3: Introducing Foundry Analytics 📊 We’re excited to roll out Foundry Analytics — real-time embedded insights built directly into Ocular Foundry to help teams optimize model performance and annotation workflows. 📉 Legacy tools fall short when it comes to data quality. They lack visibility into: - Annotation distribution - Label imbalance - Annotation consistency - Version-level dataset structure 📈 With Foundry Analytics, teams can: - Catch data issues early before model training - Track annotation progress across jobs and batches - Visualize label balance and dataset splits - Ensure high-quality, training-ready data Because data quantity helps… but data quality wins. Start building better, more reliable AI models — powered by data you actually understand. 📖 Blog: https://lnkd.in/g8w4uCPP 📅 Book a demo: https://lnkd.in/gKSA4UpT #LaunchWeek #AIAnalytics #MultimodalAI #ModelTraining #ComputerVision #DataQuality #FoundryAnalytics #OcularAI

Launch Week III, Day 2: Autonomous Data Agents for Annotation ⚡ We’re excited to ship Foundry Data Agents — autonomous agents for multimodal data annotation workflows. Manual annotation has long been a bottleneck for Computer Vision and Multimodal AI. It’s slow, repetitive, and difficult to scale. That changes today. With Foundry Data Agents, you can now: ✅ Auto-label entire jobs or specific frames ✅ Set per-label confidence scores ✅ Use our interactive playground to test before scaling ✅ Review & approve with human-in-the-loop workflows ✅ Save up to 90% of annotation time and 80% of labeling costs Powered by state-of-the-art models like GroundingDino and Grounding SAM, and built directly into an infinite canvas, Data Agents work just like Cursor does for your code — intelligent copilots, but for your data. And there’s more on the way: 🧠 Coming soon: Support for Large Language Models (GPT-4o, Gemini, IBM Granite, Claude, Llama, and more) 🔁 Hybrid workflows for continuous improvement We're making multimodal data processing as seamless and intuitive as working with structured data — so your teams can go from raw data to fine-tuned models faster than ever. Blog: https://lnkd.in/gFtQEDzv Learn more: https://lnkd.in/grk_rZRK Book a demo: https://lnkd.in/gKSA4UpT #OcularAI #YCW24

Launch Week III, Day 2: — Natural Language Search on Foundry 🔍 We’re excited to launch Foundry Multimodal Search, unlocking the fastest way to find what’s inside your video and image data using plain language. Manual tagging has long been the industry workaround — but it’s slow, error-prone, and doesn’t scale. Now, your team can simply ask: 🧍 “Show me people walking at night” 🚗 “Find red cars in parking lots” 📦 “Where’s the forklift near a loading dock?” 🔤 “Find scenes with the word ‘Caution’ on a sign” (yes, OCR is built-in) Then instantly: Surface semantic moments Jump to time-stamped clips Read AI-generated summaries Explore via interactive timeline views Search embedded text inside videos and images All of this is built on top of the Ocular Multimodal Lakehouse — launched yesterday. With Foundry Multimodal Search, teams can now: ✅ Curate training data using natural language ✅ Integrate search into internal tools and workflows with APIs ✅ Generate and store vector embeddings for reuse ✅ Power recommendations, RAG, QA, and fine-tuning pipelines 📖 Blog: https://lnkd.in/gVgJSpES 🔍 Learn more: https://lnkd.in/gavtuidw 📅 Book a Demo: https://lnkd.in/gKSA4UpT #LaunchWeek #MultimodalSearch #ComputerVision #OCR #LLMs #RAG #AIInfrastructure #OcularAI #YCW24

Launch Week III, Day 1: 🧠 The AI-Native Multimodal Data Lakehouse is here! 🚀 We’re kicking off Launch Week III with a major drop — introducing the Ocular Multimodal Data Lakehouse, built from the ground up to handle unstructured, multimodal data at scale. Modern AI models aren’t trained on rows and columns. They rely on video, images, audio, text, and other real-world data. Yet most data infrastructure today is still optimized for structured SQL-based workflows. Ocular changes that. Built as an AI-native system, the Ocular Lakehouse allows you to: ✅ Ingest multimodal batch data from cloud, local, or partner sources ✅ Collaboratively, catalog, curate, and preview raw files ✅ Build semantic indexes for labeling, fine-tuning, and evaluations ✅ Govern assets with versioning, lineage, and access control No more spreadsheet-based tracking, broken pipelines, or duplicated data. This is the missing layer for every AI team scaling multimodal workflows. Learn more: https://lnkd.in/gnnpgNfS #MultimodalAI #DataLakehouse #AIInfrastructure #ComputerVision #MLOps #LaunchWeek #OcularAI #YCW24

Launch Week III is here! We started as a simple labeling tool—now we’re a full-stack platform for multimodal AI. This week, we’re rolling out new features across Data, Search, Annotation, and Model Training (starting with CV models). This is the Multimodal AI Era!