Jominok

Inspiration

Web3 funding has led to over $1 billion in grants issued, but a lack of standardized validation means funds can be misallocated. Many projects go inactive after receiving grants, leading to inefficiencies in capital distribution. Jominok was built to solve this by automating project activity assessment and ensuring funding goes to actively building teams.

What It Does

Jominok is an AI-powered agent that:

• Monitors on-chain and off-chain activity of grant-funded projects, aligning with Bitte’s NEAR/EVM Chain Agent Bounty.
  
• Automates grant validation by assessing social media, GitHub, and transaction activity, contributing to Questflow’s Strategic Trading & Airdrop Agent Bounty.
  
• Generates AI-powered reports on project health, fitting into Aurora’s AI Agentic Chains Bounty for autonomous AI decision-making.
  

How We Built It

NEAR Protocol, OpenAI LLMs, Next.js, Rust, TypeScript, ZeroMQ, Socket.IO, Nearblocks Indexer API, Masa Node for Twitter Data, Langchain RAG for AI-assisted reporting.

Challenges We Ran Into

• Data Accuracy – Ensuring metrics reflect true project activity.
  
• On-Chain Data Processing – Balancing performance with decentralized verification, key for Nuffle Labs’ best integration challenge.
  
• Automating AI Report Generation – Making grant reports fact-based yet human-readable, a use case for Swanchain’s Marketing Agents Bounty.
  

Accomplishments That We’re Proud Of

• Developed a fully functional grant monitoring system that can be adopted by funding platforms.
  
• Integrated AI-powered scoring mechanisms for Web3 project health assessment, making it relevant for Proximity Labs’ Cross-Chain Trading Agents Bounty.
  
• Built an on-chain verification mechanism for accountability, aligning with HOT Wallet Bounty’s Omni Bridge integration.
  

What We Learned

• AI-Driven Grant Validation Works – Automating this process saves time and improves fund distribution.
  
• Cross-Chain Funding Insights Are Valuable – Enhancing grant monitoring across chains is a key goal for Proximity Labs’ Bitcoin Agents Bounty.
  
• AI-Generated Reports Can Power Web3 Governance – Improving funding decisions through AI aligns with VeaxFlow AI Agent Bounty.
  

What's Next for Jominok

🔍 AI-Enhanced Grant Criteria – Making funding decisions more data-driven.
📊 Expanded Monitoring – Covering additional Web3 ecosystems beyond NEAR.
🛠 Integration with Major Funding Platforms – Bringing Jominok’s agent to platforms like Gitcoin and DoraHacks.
🚀 AI-Powered Investment Risk Analysis – Extending to Web3 VC funding for smarter capital deployment.

Overview

The Jominok is designed to automatically assess the activity level of blockchain projects and determine if they can be considered "dead" or "inactive". The agent assesses social media scores, github scores, and on-chain activity scores to determine how active the project is building so that the analysis can be further used as a criteria for the grant allocation.

Problem

According to Gitcoin’s State of the Web3 Grant Report, there are over 1B $USD of grant issued across 5,900 projects in 2023. However, it lacks a standardized and automated measure to assess the activity of the project such that it requires extra effort for validation and the fund may be allocated to the wrong project.

Solution

Jominok offers an automated agent that

1. Collects data from the projects’ social(ex. X), building(ex. Github), and user(ex. onchain transaction) activity and store it on the activity database
2. Calculates the activity score based on the score calculator algorithm for “Dead” and “Active” analysis
3. Reports whether the project is dead or alive with the analysis data to back up the judgement. The report will be stored onchain for further verification

System Architecture

original proposition

Updated 241101

System Flow

The project consist of three main flows, which is coded alphabetically.

• Collect & organize data points: This means figuring out which project is the target for monitoring, and acquiring the project's social/github/wallet address. This part is less about programming; it's more about coordinating with funded projects for information.
• Collect activity data: Once data points are set, we use APIs to monitor and collect activity data in regular cycle.
• Judgement and action: Activity data is ingested to llm-based algorithm. Based on the result score, fact-based reports are created (by llm), actions may be called, and the info is updated to offchain DB & mainnet.

Details

• Data point / Collector: Github / X Api collecter, Near indexing
• Data Sets for the Metrics: Web3 Grant Funding DB (most likely to be Gitcoin but need discussion)
• Score Calculator: Multipath reasoning algorithm that utilizes LLM-as-a-judge
• Report: Instruction based prompt engineering with RAG on Langchain
• Report UI: Colab with metrics and instructions on how to customize them

Impact

• Historical Grant Assessment: Check and see whether previous grant projects are still in progress or not
• Grant Project Management: Check the current grant projects to see if their activity should be eligible for the amount
• Grant Criteria Setup: Set up metrics for the least / preferred required activities for the grant recipient

Built With

• aurora
• cli
• cloud
• console
• express.js
• llms
• near
• next.js
• openai
• protocol
• rust
• socket.io
• typescript
• zeromq