# DeepGov GG23

## Work Packages & Milestones


*Figure A: Defining the work packages for GG23*


# DeepGov Work Packages

## WP1: AI as the engine
This work package develops specialized AI agents that process complex ecosystem data to support human decision-making. The AI engine serves as the computational backbone of the Celo EVAL initiative.

- **M1.1 Impact Evaluation**: AI systems that assess project outcomes against measurable goals using on-chain and off-chain data, providing objective impact metrics.
*Technical considerations: Data integration pipelines connecting multi-source inputs, standardized metrics APIs, NLP models for contextual analysis, and scalable ML infrastructure for processing large datasets.*
*Deliverable D1.1.1 (March 31-April 7): Run impact evaluation on OSS projects and save results into supabase*
**[UPDATED: Timeline extended to cover March 31-April 14, with work continuing through the entire GG23 period]**

- **M1.2 Application Review**: Automated analysis of funding applications against established criteria, helping identify promising projects and standardizing evaluation processes.
*Technical considerations: Document understanding models, semantic similarity algorithms, automated risk assessment frameworks, and integration with existing application databases.*
*Deliverables:*
- *D1.2.1 (April 1-7): Run application review on OSS projects and visualize results*
- *D1.2.2 (April 7-14): Run application review pipeline on all gitcoin applications*
**[UPDATED: Timeline extended with work continuing through April 14]**

- **M1.3 Credit Assignment**: Systems that attribute ecosystem growth to specific interventions, creating clear connections between funded initiatives and measurable outcomes.
*Technical considerations: Causal inference algorithms, multi-factor attribution models, time-series analysis tooling, and transparent calculation methods with confidence scoring.*
*Deliverable D1.3.1 (April 7-21): Assign agent credit on OSS and Hypercerts round*
**[UPDATED: Timeline shifted to begin on April 14 and extend beyond GG23 voting end date]**

## WP2: Human as the steering wheel
This work package ensures human control and oversight of the AI systems, implementing Vitalik Buterin's framework where humans maintain decision authority while AI provides analysis.

- **M2.1 Avatar UI**: User interfaces that provide virtual representations of AI agents, making their capabilities and reasoning accessible to community members.
*Technical considerations: Conversational AI interfaces, explainable AI visualizations, personalized interaction patterns, and multi-device compatibility with accessibility standards.*
*Deliverable D2.1.1 (March 31-April 7): Define three AI personas, and create avatar profiles, model specs and Github repos*
**[UPDATED: Timeline extended to cover March 31-April 14]**

- **M2.2 Compass UI**: The DeepGov Compass interface that helps users find alignment between their values and governance decisions, serving as an AI-powered advisory tool.
*Technical considerations: Value-alignment algorithms, interactive preference mapping tools, real-time recommendation engines, and responsive front-end architecture.*
*Deliverable D2.2.1 (April 7-14): Deploy updated Compass UI to 'get to know' AI*
**[UPDATED: Timeline shifted to begin on April 14 and extend to April 21]**

- **M2.3 Voting UI**: Governance interfaces that integrate AI insights while preserving human decision-making in the voting process, enhancing participation through reduced information barriers.
*Technical considerations: Secure voting mechanisms with cryptographic verification, governance platform integrations, vote-weight calculation frameworks, and real-time voting analytics dashboards.*
*Deliverables:*
- *D2.3.1 (April 7): Voting period starts*
- *D2.3.2 (April 8-13): Show live voting results*
- *D2.3.3 (April 14): Aggregate all user votes and distribute X$ according to voting result*
**[UPDATED: Timeline shifted to begin on April 14 with voting period running from April 14-21]**

## Key Timeline Updates
- **Extended Development Period**: The overall timeline has been extended, with major milestones continuing through April 23 (previously concluded on April 14)
- **Staggered UI Releases**: The Avatar UI, Compass UI, and Voting UI components are now released sequentially rather than concurrently
- **Retro Announcement**: Added new milestone "Announcement synced with Retro" aligned with the end of Retro period (April 23)
- **Voting Period**: Moved from April 7-14 to April 14-21, giving more time for development and testing before user voting begins
- **First Reviews and DeepGov Release**: Additional milestone markers added for "First Reviews" (April 7) and "Broad first release" (April 14)

---

## Goals
What goals do we have with the GG23? What value are we creating for Gitcoin?

- automate application reviews
- makes life easier for application reviewers
- could be a way to increase quality of reviews (I've seen complaints about reviewers missing basic things for Cabin DAO)
- demonstrate AI agents (AI Engine + Human Steering / DeepGov)
- innovative way to vote/donate to projects
- opening a new frontier of mechanism designs
- Gitcoin could be early adopters of this new cycle/trend

## Deliverables
What deliverables? What are we shipping? Same as for Celo with some improvements + additions.

### AI Agents
- Create Gitcoin AI Agent Personalities

### Compass
- Adjust questions to align with AI agents

### Voting UI
Connect each voter to an address
- Connect Wallet
- Sign message + Verify signature before storing vote data
- Some kind of sybil defense (passport to filter approved votes?)
- *(can be added later)*

### AI Avatar updates?
- Update the avatars

### Model Specs updates?
- Have a temperature check on Gitcoin community?

### Agent Framework

API for calling the Agents to review the application, storing the results in EAS or Web2 db. (EAS makes it better interoperable with other systems and more decentralized but adds a transaction cost to store the data)

### Application Review Dashboard

Applications (or Gov Proposals) can be ingested and reviewed by 5 AI Agents. These applications will be displayed in a list with a link to navigate to Application Details. This page shows the details extracted from the application (project description, etc). Each agent reviews and scores the application with a Summary, Strength & Weaknesses, and Recommendations. The review includes sources it found during its research.

- Ingest Applications (only admin)
- Call API with application data (+ context such as round information)
- Call 5 Agents to Review each Application (Agent Framework)
- System Prompt + Model Spec + Round Context + Application
- List Applications (each application will have 5 reviews)
- Table to list all reviewed applications and their scores
- View Application Details
- Show Project name + description
- List Reviews for Application (5 Agents)
- Each Review has a score, summary/feedback/reasoning, strength & weaknesses, requested changes



### Flows

sequenceDiagram
actor voter as Voter
participant v as Voting UI
participant gg as GG23

note over voter, gg: Voter has some amount of token from the beginning

### Sense-Making

Sketch out prompts here for how we want the inputs and outputs to the LLM. For this, we started this Polis questionaire seeding it with 10 questions: https://pol.is/9my4ftmrim

### Phase 1: AI Agents

We have defined three personas that aim to represent three distinct sub communities within Gitcoin (Regens, Technocratic/Liberals, Pluralists/Left)

#### Panda the Regeneratoor

#### Luna the Open-Source Capitalist

#### Woki the Gitcoin Commie

### Phase 2: Impact Evaluation


### Phase 3: Election Campaign