The problem datalatte solves

In the digital age, our online interactions generate vast amounts of data often controlled and exploited by centralized entities, leading to issues of control and transparency. Additionally, LLM-based applications & agents face a cold start problem, lacking initial user data for personalization.

Datalatte addresses these challenges by allowing users to reclaim ownership and control over their data through a gamified experience. Users' data is stored in a personal knowledge graph (semantic layer), accessible by their AI agents on any platform with user authorization.

Key Features

• Ownership and Control: Datalatte enables users to build a digital twin from their onchain and offchain data, giving them full control over access and purpose.
• Consent-Based Access: Users can specify access terms for their data, ensuring use is only with their explicit consent, allowing for data monetization on their terms.
• Semantic data layer: Storing data in knowledge graph data models provides an efficient & accurate retrieval of information while enabling LLM-based agents to perform better in reasoning and context understanding of the tasks as personal AI agents.

Use Cases

• Personalized Experiences: Third-party applications and agents can access users' digital twins to personalize experiences while maintaining user control and privacy.
• Data Monetization: Users can monetize their data, generating income from their digital activities.
• Familiarizing with AI Agents: A gaming environment helps users understand AI agents, their use cases, and benefits, reducing hesitation in trusting onchain transactions conducted by agents.

Challenges we ran into

Gating the NFT Mint for Authentic Users: In a dataDAO, ensuring authentic users with genuine data is crucial. To protect our NFT drop from spam, bots, and non-cohesive profiles, we use several whitelisting methods. Eligibility includes holders of Power Badges, accounts liked by the @notabot account (via Airstack), and the top 10k FIDs in global rank via the OpenRank API. Users outside these categories can inquire directly and may need to achieve a minimum of 24 points on passport.gitcoin.co. Our LLM-based pipeline checks for authentic and cohesive behavior, prioritizing genuine self-expression over engagement farming.

Ranking Authentic Users: Only the top 1010 users can mint a dataBarista NFT. Users receive a digital twin score based on personal attributes revealed in their cast history. By visiting our frame daily, users can earn up to 25 scores based on the authenticity of their posts, favoring genuine self-expression over farming and spamming activities.

Pipelines: We use the crewAI agent framework with multiple agents working sequentially to filter cast history and extract relevant data (attribute scores, roles/skills, and inventory items) for the gaming character. We also extract meaningful features for users' personal knowledge graphs, linked to their FID. Google Gemini's long context window (1M tokens) processes entire cast histories to extract valuable information.

Ensuring a Smooth Frame Experience: Fetching the entire Farcaster cast history for some accounts can be time-consuming. We have gained access to run direct SQL queries on Neynar's hosted hubs, allowing us to retrieve the last 2500 casts of each user efficiently.

Note Regarding Our GitHub Repo: Only the frame is public. All backend services, including pipelines and scoring criteria for authentic users, remain private to prevent bots from emulating favorable actions.