gbrllim/openrefinery.ai

Decentralised Crowdsourcing Platform built on ICP

OpenRefinery.AI - Add your human touch to tomorrow's AI Chatbots

Live deployment link: https://vzxr6-nqaaa-aaaal-aja7q-cai.icp0.io/

(Please connect with an internet identity wallet for the best experience)

Background

In today’s rapidly evolving landscape, artificial intelligence (AI) has become an integral component across diverse industries. However, the efficacy of AI models hinges largely on the quality of their training datasets. A good training dataset will ensure that the model can produce accurate, reliable results and reduce training costs. This is why the training dataset market is projected to grow 24.16% annually.

Chatbots are instrumental in providing personalised interactions, which enhances customer engagement, retention and ultimately, sales. 62% of consumers prefer interacting with chatbots over human agents, and consumer spending via chatbots in the retail industry is projected to soar to USD 158 billion by 2025.

Whilst there are existing solutions (MTurk, CrowdFlower, Clickworker, etc), they are hampered by transparency issues, quality control concerns, lack of decentralisation and multilingual support challenges.

What is OpenRefinery.AI

Our vision: Humanising and refining data on the blockchain to empower AI

Our mission: The leading platform in decentralized and transparent data crowdsourcing. We aim to produce top-tier training data, driving AI technology forward and enabling smooth human-AI interaction

OpenRefinery.AI is a revolutionary crowdsourcing platform that redefines the AI training landscape. We ensure AI understands human communications nuances completely and reaches its 100% potential by enabling it to comprehend diverse cultural backgrounds and languages seamlessly.

Why ICP

• Internet Computer Protocol (ICP) operates as a self-running cloud to the internet using "canister software" on a decentralised network
• 100% of the online services are functioning on the blockchain for complete decentralisation and cost-effectiveness
• ICP is designed to scale horizontally, allowing applications to handle increasing traffic and loads without sacrificing security or performance
• Cost of computation is put on the developers, making it affordable for users to interact with applications without having to incur gas fees with every transaction

User Roles & Key Terms

• Creator: Lists and sponsors projects that require high quality datasets to be generated for specific topics
• Miner: Generates original paraphrases for the subjects listed. Compensation is provided only after successful validation by an Inspector
• Inspector: Ensures the quality and originality of the paraphrases to maintain the integrity of the data set
• Project: A set of requirements and subjects established by a Creator that requires crowdsourced effort to generate relevant paraphrases.
• Subject: A topic sentence that requires additional data (e.g., "I need to reset my password", "I want to change my profile picture").
• Paraphrase: A sentence similar to the subject but rephrased (e.g., "How can I update my password?", "How do I change my display photo?").

Core Features

• Project Management:
  • Flexible Project Creation: Allow for a variety of tasks, project creators can indicate their preferred payout, paraphrases needed, inspection count and language needed for more complex projects
  • Examples galore: Miners are provided with live updates on all paraphrases currently pending, approved and rejected. This prevents duplicate work and encourages high quality paraphrases
  • Progress tracking: Track the progress of your projects live using our dashboard and analysis tools
• Decentralised Infrastructure:
  • Web3 Authentication: Create account anonymously using Internet Identity
  • Decentralised Storage: Store images and data entirely on-chain
  • Gasless transactions: Users are able to transact and interact with the blockchain without having to incur gas fees
• Validation Mechanism:
  • Consensus Handling: All paraphrases generated require a strict 80% approval rate to be passed, ensuring majority consensus
  • Approval Ratings: All users are tracked with an approval score to sure quality validation of paraphrases thereby deterring bad actors
  • GemsGemsGems: In-house token reward system to facilitate incentives and ease of payments to Miners & Inspectors

Roadmap

• Reputation and review system - track performance scores based on ratings
• KYC Onboarding for new users
• Referral rewards
• Milestone reward targets (100, 1000, 10000 tasks completed)
• Audio, Image and Video annotaion support
• Survey data colleciton
• Demographic targeting & data collection
• Mobile application on iOS and Android

Technical Stack

• Juno Build
• Internet Identity (ICP)
• ReactJS/ViteJS
• TailwindCSS
• React-router

Contributors

• Vanessa Thien - Leader, UIUX
• Akshaya Arumugham - Research
• Kwang Guo Dong - Product
• Gabriel Lim - Engineering

Launching

To get started, clone the repository and run the following commands in the project directory.

OpenRefinery is best run on a web browser, we recommend Brave/Chrome :)

1. Run npm i to install NPM packages
2. Then npm run dev to run the app in development mode
3. Open up http://localhost:5173 to view it in your browser
4. Connect with Internet Identity for the best experience