Recap 2024 - An Epic Foundational Year 

That’s a wrap for 2024, and what an amazing journey it has been helping our customers extract and use meaningful business context from their unstructured data in the easiest way possible. At Encord, we strive to be the last AI data platform teams will need to efficiently discover and prepare high-quality, relevant private datasets for training and fine-tuning AI models at scale.  

Encord customers are pushing the boundaries on how AI can help improve business operations, save lives, delight users and customers, and, most importantly, make GenAI and custom models work better for businesses with richer data. All this while being maniacal about our customer experiences and building a lasting AI company. 

This year we’ve:

1. Helped customers like Synthesia and Flawless AI achieve groundbreaking GenAI research.
2. Onboarded AI innovators like 
3. Showed the world that multimodal is possible in a unified AI data platform while releasing  ___ game-changing and foundational product enhancements, including support for SAM 2 within 48 hrs of its public release. 
4. Closed our $32M Series B to further support R&D and GTM
5. Opened our San Francisco office to build and scale our global GTM functions.

In addition to delighting our customers, in 2024, we evolved our industry-leading computer vision and medical AI data platform to enable teams to easily discover, manage, curate, and annotate petabyte scale document, text, and audio datasets. We also introduced a multimodal annotation interface facilitating reinforced learning from human feedback (RLHF) workflows and multi-file analysis and annotation in one view. Teams can now view video, audio, text and DICOM files in one interface to seamlessly orchestrate multimodal data workflows, fully customizable for any use case or project. What does this all mean, we are finishing 2024 as the only end-to-end AI data platform for multimodal data.

Teams building AI systems for Computer Vision, Predictive, Generative, Conversational, and Physical AI can now also use Encord to efficiently transform petabytes of unstructured multimodal data into high-quality, representative datasets for training, fine-tuning and aligning AI models. Let's recap the highlights that our customers loved most. 

Audio

Encord’s audio data curation and annotation capability is specifically designed to enable effective annotation workflows for AI teams working with any type and size of audio dataset, literally any size. Teams can accurately classify multiple attributes within the same audio file with extreme precision down to the millisecond using customizable hotkeys or the intuitive user interface.  Whether you are building models for speech recognition, sound classification, or sentiment analysis for your contact center workflows, Encord provides a flexible, user-friendly platform to accommodate any audio and multimodal AI project regardless of complexity or size.

Documents and Text

AI Teams can use Encord for any annotation use case to comprehensively and accurately label large-scale document and text datasets, including: Named Entity Recognition (NER), Sentiment Analysis, Text Classification, Translation, Summarization, and RLHF.  Comprehensive annotation and quality control capabilities include the following:

• Customizable hotkeys and intuitive text highlighting - speeds up annotation workflows.
• Pagination navigation - whole documents can be viewed and annotated in a single task interface allowing for seamless navigation between pages for analysis and labeling.
• Flexible bounding box tools - teams can annotate multimodal content such as images, graphs and other information types within a document using bounding boxes.
• Free-form text labels - flexible commenting functionality to annotate keywords and text and the ability to add general comments.
  

Multimodal Annotation

Using the customizable multimodal annotation interface, teams can now view, analyze, and annotate multimodal files in one interface.  This unlocks a variety of cases that previously were only possible through cumbersome workarounds, including:

• Analyzing PDF reports alongside images, videos, or DICOM files to improve the accuracy and efficiency of annotation workflows by empowering labelers with extreme context.  
• Orchestrating RLHF workflows to compare and rank GenAI model outputs such as video, audio, and text content.  
• Annotate multiple videos or images showing different views of the same event. 

Encord customers have already saved hours by eliminating the process of manually stitching video and image data together for same-scenario analysis. Instead, they now use Encord’s multimodal annotation interface to automatically achieve the correct layout required for multi-file annotation in one view.

Data Agents

Earlier this year, we also released Encord Data Agents, which enable teams to integrate AI models into their data workflows in a highly customizable way. Teams have integrated their own or foundation models, such as OpenAI’s GPT-4o and Anthropic’s Claude 3 Opus, to pre-label large datasets and smart-routing within data workflows and auto-reviews.  

Using Encord Agents, teams are saving __ annotation time, boosting label throughput, and finding more label errors per expert review hour through agent integrations of both foundation models and in-house models.

Teams can use the Encord Agents Library, a powerful yet flexible and lightweight framework that abstracts all the details of platform integration to integrate models into data workflows even faster. 

The Encord Agents Library enables:

• Seamless access to the data and labels you need in a simple, accessible API.  
• Shorter time-to-value, allowing you to build and run Agents in a matter of minutes instead of hours.
• With APIs for Editor and Task Agents and one-line CLI test commands, you can prototype, build, and integrate cutting-edge models into your workflows easier than ever.

SAM 2 for Accelerated Data Annotation

Meta released Segment Anything Model 2 in July, and within 48 hrs of its release, Encord customers were able to leverage SAM 2 natively within the Encord platform to improve and accelerate mask prediction and object segmentation in image and video data.  Our customers have used the model millions of times to automate their labeling processes and have seen huge benefits of 6x faster performance compared to the original SAM model. Accessing SAM 2 capabilities natively in Encord has also saved AI teams hours of time and manual effort by eliminating the need to label individual frames of video for complex object masking. 

Data Curation and Management

Over the past few years, we have been working with some of the world’s leading AI teams at Synthesia, Philips, and Tractable to provide world-class infrastructure for data-centric AI development.  In conversations with many of our customers, we discovered a common pattern: teams have petabytes of data scattered across multiple cloud and on-premise data storages, leading to poor data management and curation. 

Enter Encord Index. Index enables AI teams to unify massive datasets across countless distributed sources to securely discover, manage, and visualize billions of data files on one platform. By simply connecting cloud or on-prem data stores via our API or using our SDK, teams can instantly manage and visualize all of their unstructured data on Index. This view is dynamic and includes any new data that organizations accumulate following initial setup. 

Teams can use granular data exploration functionality within to discover, visualize, and organize the full spectrum of real-world business data and a range of edge cases:

• Embeddings plots to visualize and understand large-scale datasets in seconds and curate the right data for downstream data workflows.
• Automatic error detection helps surface duplicates or corrupt files to automate data cleansing.  
• Powerful natural language search capabilities empower data teams to automatically find the right data in seconds, eliminating the need to manually sort through folders of irrelevant data. 
• Metadata filtering allows teams to find the data that they already know will be the most valuable addition to your datasets.

As a result, our customers have achieved, on average, a 35% reduction in dataset size by curating the best data, seen upwards of 20% improvement in model performance, and saved hundreds of thousands of dollars in compute and human annotation costs. 

We’re just getting started

Encord is designed to enable teams to future-proof their data pipelines for growth in any direction—whether they are advancing laterally from unimodal to multimodal model development or looking for a secure platform to handle rapidly evolving datasets at petabyte scale. 

Encord unites AI, data science, machine learning, and data engineering teams with a consolidated platform to search, curate, and label unstructured data, including images, videos, audio files, documents, and DICOM files, into the high-quality data needed to deliver improved model performance and production AI models faster.

Our customers' focus on democratizing AI across businesses everywhere, paired with our relentless drive to delight our customers with magical product experiences, is the perfect foundation for an even more exciting 2025!