Fund Public Goods with AI

Public goods are important. Projects like Gitcoin, Optimism, Octant, DAO Drops, and Giveth are part of a movement that reimagines the way public goods are funded. These ecosystems utilize a variety of mechanisms to help participants fund what matters in a manner that is transparent and bottom-up.

But these platforms share a common problem: it’s difficult to surface the most impactful projects and decide how much to fund them. And it’s especially hard as the diversity and number of projects increases over time. Perhaps AI agents can help.

Streamlining the fund allocation process with agents

In many DAOs, blockchain foundations, and public goods funding platforms there are hundreds of projects and individuals seeking funding. As this number grows, it isn’t feasible for decision-makers to research, evaluate and decide on the optimal level of support for each option. This is where AI agents come in. Within minutes, agents can scrape the web, social media, and other data sources to compile and synthesize relevant information across a large number of projects.

Based on these findings, the agents can then decide on a funding allocation across projects and prepare the associated transaction for the user. The user could be an individual donor, a DAO allocating contributor rewards, or even a government. Agents can not only make the funding process more efficient, but can also reduce bias and increase the ability to customize funding allocations to user preference.

Fundpublicgoods.ai is the first AI agent designed to help streamline the public goods funding process. It focuses on helping individuals make highly informed donation decisions.

How it works

The agent’s research pipeline is designed to generate custom public goods allocation strategies based on the user’s preferences and publicly available data. Here are the steps the agent takes:

1. Find Relevant Projects. Perform a vector similarity search across project data from all past Gitcoin rounds. Aggregate the results, filter out duplicates, and rerank the results based on the original prompt to arrive at the top ten. Techniques used: Similarity Search, LLM-based Reranking
2. Generate Holistic Evaluations. Generate a comprehensive report for each project that breaks down its a) relevance, b) impact, and c) funding needs. Techniques used: Generated Knowledge Prompting, Zero-Shot Prompting, Summarization
3. Compute Evaluation Scores. Analyze each report and assign a numerical evaluation score from 0-10 for each of the three criteria. Combine these scores through a weighted average to arrive at the project’s overall score. Weight the allocation percentages in proportion to scores of the projects with addresses on the selected chain. Techniques used: RAG, Numerical Evaluations, Algorithmic Composition

The allocation strategy includes a report for each project summarizing the agent’s findings.

One critical aspect of the report is the impact section. The agent uses information from the project’s Gitcoin round applications to arrive at an intelligent estimate of the impact that a project has had.

Another unique aspect of the report is that it estimates each project’s funding needs based on information like prior funding and team size. This help gives context on the project’s past funding sources, such as venture backing or grants.

An important limitation to note is that this version of the agent only uses self-reported data from projects. But future versions could use a variety of on- and off-chain data from third parties.

Try the agent today

Fundpublicgoods.ai is now live. Try using it to donate to your favorite public goods projects and let us know what you think on Discord.

What’s next for the PGF agent?

The PGF agent marks the beginning of an exciting journey into the convergence of AI agents and web3 protocols. This experiment in AI-driven public goods showcases how agents can reshape web3 experiences. To truly harness the transformative power of agents in this domain, several key enhancements are being considered for integration into future iterations:

• Advanced Data Integration: Expand the data sources to include a wider range of information, such as social media sentiment analysis, metrics from Open Source Observer, attestations from Ethereum Attestation Service, and impact certificates from Hypercerts.
• Personalized Impact Metrics: Allow users to customize impact evaluations based on their values and preferences. The agent would tailor its impact analysis based on the types of impact that the user cares about most.
• Automated Allocations: Enable users to create a "set-and-forget" allocation strategy, ensuring ongoing support for their favorite projects without needing to manually initiate transactions.
• Enhanced Scoring Techniques: Improve the evaluation process with more sophisticated scoring algorithms. For example, deriving the numerical evaluations from a detailed set of yes/no questions posed to the LLM about project data rather than asking the LLM for a score directly.
• Enhanced Search Techniques: Improve relevance of results and reduce load time by implementing additional techniques like HyDE, Query Transformation, and Hybrid Search.

We look forward to working with different web3 ecosystems to build new versions of the agent with various enhancements that support their public goods funding needs.

Expanding Web3 horizons with AI agents

Beyond public goods funding, Agentcoin is applying its AI agent tech to a range of new agent-mediated web3 experiences. Agents can not only make web3 more accessible, but also lay the groundwork for unlocking web3 use-cases that require scalable intelligence – from impact evaluation and risk analysis to governance, dispute resolution, prediction markets and more.

You are invited to join the Agentcoin community in reimagining the web3 experience using AI agents on the Agentcoin Discord.