Annotat3

• 

  Home page featuring trending dataset and most active users

• 

  Platform stats page

• 

  Datasets Library with filtering options

• 

  Annotation task module

• 

  Personal profile page tracking rewards and tasks

• 

  High level diagram of the solution

Inspiration

The primary catalyst behind Annotat3 was addressing two significant hurdles prevalent in the field of Large Language Models (LLMs) and other AI modalities: the lack of transparency in training data and the high costs linked with the laborious process of labeling efficient datasets.

Functionality

Annotat3 is a unique platform that enables users to upload and annotate datasets. Upon the upload of a dataset, a ComposeDB model is crafted specifically for that data. Concurrently, a data Non-Fungible Token (NFT) is minted using the Ocean Protocol, and a splitter contract for rewards allocation is created.

When a user opts to annotate this data, the annotations are encrypted and stored along with the annotated data in a ComposeDB annotation model. To determine the most accurate annotation based on multiple inputs, each image hash annotation is projected into an embedding space to locate the central point, or the core of the annotation cluster.

Users interested in leveraging the annotated data can purchase access to the data NFT symbolizing the annotated dataset. The ensuing proceeds are secured in a Safe contract, which then employs the splitter contract to distribute rewards to the annotators. The rewards are proportional to the annotators' proximity to the clustered point in the embedding space, with those closest receiving the majority of the rewards.

Development Process

We crafted the front end using Next.js combined with Tailwind CSS. For encryption and storage, we relied on Lit Protocol and ComposeDB respectively. The data market and authentication process were facilitated by the Ocean Protocol, while smart contracts were used for fund distribution.

Encountered Challenges

Two key challenges were faced during the development process. Firstly, working with ComposeDB composites and identifying the right data structure proved to be an intricate task. Secondly, we experienced difficulties while integrating a UI framework with Tailwind.

Our Proud Achievements

We're particularly proud of successfully implementing the Lit Protocol and ComposeDB. Besides, we managed to develop an efficient scoring algorithm that effectively ranks annotators and delivers high-quality data without needing a centralized assessor.

Acquired Knowledge

Our journey with Annotat3 has been insightful as we delved into the Ocean Protocol and its structure around data NFTs. We also expanded our knowledge about Bacaleau and Feltlabs' usage for enabling Compute over data and federated learning.

Future of Annotat3

The next steps for Annotat3 encompass the development of a fully functional application that is user-friendly. We envision this platform as a cornerstone in shaping the democratization of data annotation.

Built With

• composedb
• javascript
• litprotocol