project screenshot 1
project screenshot 2
project screenshot 3
project screenshot 4

Deano

A Decentralized Data Annotation Platform leverages blockchain technology to revolutionize data labeling and annotation. It distributes tasks across a network, ensuring transparency, data privacy, and high quality.

Deano

Created At

ETHOnline 2023

Winner of

🌟 Sismo — Most Creative

🥉 Tableland — Best Use of Studio

🥇 XMTP — Best Use

Project Description

Deano - A decentralised annotation platform

Introduction

Authenticity and Quality of Data used for training Machine Learning models is a major concern in the industry. We often have to blindly trust the data provided by centralized data providers and there is no way to verify the authenticity of the data.

Deano tries to solve this problem by providing a transparent and decentralized platform for data annotation. Vendors can upload their data to the platform and request for annotations. Each annotation request is verified by the community and the data is annotated by the community. Each member is given a fixed payout every month for their contribution to the platform based on a reputation metric for each individual.

Project Idea

  • Authenticity and Quality of Data used for training Machine Learning models is a major concern in the industry. We often have to blindly trust the data provided by centralized data providers, and there is no way to verify the authenticity of the data.

  • The existing data annotation platforms, such as 'Kaggle,' suffer from centralization, where the data provided is often not rigorously validated for accuracy, leading to potential data quality issues. Moreover, these platforms lack a systematic reward mechanism for annotators, making it less appealing for individuals looking to contribute.

  • In response to this challenge, Deano presents a groundbreaking solution in the realm of data annotation, designed to address critical issues in the current landscape. It endeavors to create a transparent and decentralized platform that fundamentally transforms the data annotation process.

  • Deano steps in as a transformative force to rectify these shortcomings. It establishes a decentralized ecosystem where vendors can post data annotation jobs with confidence. Annotators, part of the Deano community, are incentivized with DAN tokens for accurate data labeling, creating a win-win situation for both vendors and annotators. This reputation-based reward system ensures quality and trust in the data annotation process.

In summary, Deano aims to ensure the authenticity and quality of data used for machine learning models by offering a transparent and decentralized platform. It leverages a reputation-based reward system to encourage community-driven, accurate data labeling, thereby addressing the centralization and quality issues prevalent in current data annotation platforms.

Our Approach

Actors

  • Vendors - Vendors are the people who upload their data to the platform and request for annotations.
  • Annotators - Annotators are the people who annotate the data uploaded by the vendors.

Workflow

  • Vendors upload their data to the platform and request for annotations.
  • Annotators verify the authenticity of the data and annotate the data.
  • Annotators are rewarded for their contribution to the platform based on a reputation metric for each individual.

How Accuracy is maintained

The accuracy of the annotations is maintained by a reputation aggregation system. Each annotator has a reputation score which is calculated based on the accuracy of the annotations done by the annotator.

Multiple annotators annotate the same data. These annotations are aggregated and checkout for majority vote. The annotators who have the same annotations as the majority are rewarded with a reputation score. The annotators who have different annotations are penalized with a negative reputation score.

By Game Theory, we can prove that we can go forward with this optimistic assumption that the majority of the annotators will be honest and will annotate the data correctly and there will be a few annotators who will try to cheat the system due to the design of the reputation system.

How Annotators are rewarded

Each annotator will be rewarded based on the reputation score. The reputation score is calculated based on the accuracy of the annotations done by the annotator. The reputation score is calculated using the following formula.

Reputation Score = (Number of correct annotations - Number of incorrect annotations) / (Number of correct annotations + Number of incorrect annotations)

This score is continuously updated as the annotator annotates more data. The reputation score is used to calculate the payout for each annotator. The payout is calculated using the following formula.

Payout = Reputation Score * Base Payout

How it's Made

How our platform works

Actors

  • Vendors - Vendors are the people who upload their data to the platform and request for annotations.
  • Annotators - Annotators are the people who annotate the data uploaded by the vendors.

Workflow

1 - Vendors upload their data to the platform and request for annotations. 2 - Annotators verify the authenticity of the data and annotate the data. 3 - Annotators are rewarded for their contribution to the platform based on a reputation metric for each individual. ``

Challenges We Faced

  • Implementing onchain testing for Sismo proved to be a challenging task, as tracking onchain changes was intricate.
  • While working with XMTP Chat, we faced difficulties adapting to the latest features due to breaking changes in the latest versions. Fortunately, we received valuable support from the community in resolving these conflicts.
  • Configuring and setting up the workflow for Tableland initially presented challenges. However, with substantial community assistance, we successfully addressed these issues, ensuring its smooth operation.

Technologies We Used

Nextjs TailwindCSS

Sismo

  • Sismo plays a pivotal role in providing periodic salaries to our annotators based on their reputation metrics.
  • It offers the capability to selectively distribute payouts exclusively within the Deano platform, ensuring financial transactions remain within our ecosystem.
  • Use of custom data group using REST data provider, vaultId, and on-chain zk-proof verification enabled us to build a robust reputation based periodic payout system.

Push Protocol

  • The Push Protocol serves the essential function of notifying annotators about newly added job listings by vendors.
  • It plays a crucial role in keeping the entire ecosystem up-to-date by delivering notifications that are vital to annotators' work since the platform is decentralised and need of notification and updates is vital.

Tableland

  • Tableland is a crucial component of Deano's infrastructure, serving as a comprehensive solution for the streamlined management of vendor job listings. It acts as the backbone of all the datasets within Deano, streamlining the web3 layer on top of the storage.
  • By simplifying the overall workflow within the web3 framework, Tableland enhances efficiency and accessibility, making it easier for vendors and annotators to manage and access job listings and related data.
  • In particular the batch queries feature of tableland was really useful for doing a lot of queries in a single network call reducing onchain interactions and gas costs.

XMTP Chat

  • Stay seamlessly connected with the Vendor for real-time updates on job listings and address any related queries.
  • Annotators have the convenience of direct communication with vendors for additional job-related updates.
  • In the future, our plans include implementing the Connext Network to facilitate a seamless token transfer bridge, enabling efficient transactions between vendors and annotators, as well as among annotators themselves.

Privy

  • Privy enhances the user experience of web3 products to a level comparable to web2, offering features like seamless email and social media sign-ins. It not only boosts conversion rates but also broadens the potential user base for your app.
  • Deano leverages Privy for smooth user onboarding, allowing users to authenticate using their email address and social logins. This streamlined process simplifies their entry into the web3 world.

web3.storage

  • web3.storage occupies a central role in our infrastructure, serving as the repository for both images and their associated information uploaded by vendors on the blockchain. This crucial function ensures secure and reliable data storage within the Deano platform.

The Graph

  • Our utilization of The Graph extends to the development of a dynamic leaderboard system. This system ranks the most distinguished annotators and vendors based on their reputation and frequency of usage respectively, providing a transparent and insightful view of their contributions to the Deano platform.

Heres our graph studio link: https://thegraph.com/studio/subgraph/deano-ranking/playground/

Architecture

You can checkout architecture of our platform here

background image mobile

Join the mailing list

Get the latest news and updates