TagFi
Bringing social and gamified approach to collect human classfied data at scale from verified users
Demo
About
Why are you participating for Based India? Yuga's tweet - [https://x.com/yugacohler/status/1823087061357793616](https://x.com/yugacohler/status/1823087061357793616) inspired us about the concept of an AI Agent wallet. We wanted to apply the concept of data collection and payment using agents to build **TagFi**, a TikTok reel-like app for data collection and rewards. ### Connection to the Region We are based out of India. The ancient roots of India take us to the story of **Mata Shabri**, who filtered out quality fruits for **Lord Rama**, and that’s where we thought AI needs many modern-day Shabris to generate **classified high-quality data**.
How it's made
What challenges are you focusing on? **Epoch 2024** research report and podcast from **Mustafa Soleman**, Microsoft AI CEO, indicated that by 2026, high-quality text data for training LLMs will be exhausted, a trend that could slow down AI progress. Human-classified data could be a great way to continue feeding in high-quality data for specific cases where synthetic data may not work. To make this happen, we need to address two key challenges: 1. ✅ **Verified humans** to avoid toxicity 2. 🌍 **Solving this at scale** And that’s where **TagFi** comes into the picture! How does your submission address this challenge? We address the above two challenges in the following ways: 1) **Verify users and check eligibility against user reputation**: From basic user verification systems to resist bots and bring unique users, to verifying users for their reputation in specific skills that qualify them to input data. For example, user-generated content (UGC) across different focus areas such as specific skills like medical surgery, experience like combat, and domain knowledge in wildlife. 2) **Solving at scale**: There are already several projects enabling the data economy that could address similar challenges, but we feel adding social engagement with a TikTok or Instagram-like reel-making, gamified experience with filters and rewards could bring larger user bases to generate data at volume across categories. 🔖 **Project TagFi**: Instagram / TikTok reel-like verified user input collection to drive the next generation of AI model training. **Personas include**: - Data consumer (AI companies) - Data classifier and content creator **Key Modules**: - User verification leveraging **Privado iD** - AI agent-based rewards using **Coinbase SDK** - Marketplace for data consumers to set requests - User can download the app on ios and android 🦾 **User flows** - **Simple classification flow**: A user undergoes verification and creates a profile on TagFi -> Accesses quests from the marketplace -> Classifies images across categories -> Gets rewarded - **Content generation flow**: A user undergoes verification and creates a profile on TagFi -> Accesses quests from the marketplace -> Leverages several ways to add classified data to the dataset (e.g., mobile camera, filters, etc.) -> Data could be verified by other users based on data consumer rules -> Rewards will be proportionate to the user’s reputation 🚀 **TagFi Impact**: Access to verified user generated and proprietary content will enable hot licensing and acquisition target to help further development of AI Challenges we ran into **Identification of Data Consumers**: Studied and interacted with several AI data-providing projects like **Masa**, **Vana**, **Nukl**, and others to understand demand and supply gaps in data consumption and generation. **Exploring ways to solve RLFH** at scale with: - **User verification** - **User reputation** - Establishing **review and rating** mechanisms - **Proportional incentive mechanics**