Please describe your proposed solution

As previous iterations of Catalyst have shown, voters cannot always predict which new proposals can “make a difference” and have consistently put more and more weight on established teams. DReps are supposed to change this, as Treasury disbursement will be voted on by them after the coming second phase of the Volatire Era. DReps can add a layer of qualification: the more remarkable and knowledgable a DRep, the more people will delegate to them. However, this may only work on paper – in reality a lot of funding will go to influencers or otherwise “follow the money” – i.e. the rich get richer. How can we empower DReps to truly represent the values of the community, the Cardano ethos and the collective sentiment of the community? There is the safeguarding against Sybil attacks and other forms of collusion.

Enter the DRep AI Assistant: connected to community sentiment, empowered with the foundations of Cardano, its roadmap, decentralized contributions and a well-stocked and diverse library of political science, philosophy and theory of governance and the entirety of entities contributing to Cardano and all data from previous Catalyst funds.

Our innovative AI decision making solutions are already live and being used by paid B2B customers. and Singaporean students who are using our platform to support their learning collaboration efforts.

Here are a few key metrics that show our traction so far:

• Customer Adoption: We have onboarded several commodity trading houses and as part of the product developed provided for free to students who are using the platform actively.
• Open Source transition: This product represents a Cardano-based fork of several platform and agent components, making it accessible for open-source development and collaboration. This is a big step toward creating lasting value in the decentralized space.

This collobarative online swarming game to converge on decisions is powered by some of te same agents that will be open sourced for this proposal.

The DRep AI Assistant will provide an invaluable resource to ensure that on-chain actions align with the Cardano constitution. It can provide busy DReps with access to instant, on-demand analysis and coverage of topics such as self-sovereign identity, universal basic income, controlling inflation and tweaking network parameters, understanding needs of stake pool operators and developers to name a few.

<u>Values, government and committee</u>

What are the values and roadmap of Cardano, the cypherpunk and web3 movements, and the philosophical giants on which shoulders they stand? Important thinkers on social contract, fairness, decentralization, democracy and pluralism, threats of regulation and nepotism, models of voting and better governance, self-sovereign identity, proof of stake, Sybil resistance safeguards and trade-offs like Nakamoto Coefficient vs scalability, peer-to-peer vs mass adoption etc.

<u>Funding and Catalyst</u>

Consider whether an idea was previously funded, what was the traction, what could be the factors that it did not reach its promise.

Assess which teams received funding, and how much of the tooling created or community built still in use.

Ideas from other blockchains, hackathons, funded startups: Identify bad actors and exploitative behavior in governance proposals and funding requests. Support innovative ideas and those worthy of new iterations or larger grants.

Currently running a trial with a relatively simple Llama-2-7b-chat-hf model and some 10 GB of telegram and gitbook collected text from Catalyst, our Fund-11 "Data-Driven Catalyst" LLM has provided proof-of-concept for a Cardano-, Catalyst- and liquid democracy centric Assistant.

Alignment with Cardano Values

The agent will be trained on relevant community contributions and discussions, IOHK documentation, resources like past Catalyst funds, Forum, research articles and core Cardano principles:

• Decentralization: Empowering individual DReps with accessible, high-quality information.
• Transparency: Enhancing understanding of governance processes and decisions.
• Community-Driven: Incorporating community sentiment and historical data into the decision-making process.
• Peer-reviewed Research Innovation: Leveraging cutting-edge data science and AI technology to solve governance challenges.

Our values deserve their own dedicated datasets, libraries and AI "personality". We are not well served with big tech Generative AI solutions that carry the biases of the TradFi and top-down, centralized world we want to change.

Please define the positive impact your project will have on the wider Cardano community

Giving the emerging first cohort of DReps and their growing role in on-chain governance and grants or funding decisions, the DRep AI-Assistant will have far-reaching positive impacts on the Cardano ecosystem. This may start relatively modestly, similar to the role itself, and gain emergent capabilities and a comprehensive body of knowlegde over time. Below is a list of the key benefits.

Enhanced Governance Quality: By providing DReps with comprehensive, unbiased information and analysis, the AI-Assistant will contribute to more informed decision-making, leading to better governance outcomes for the entire Cardano community.

Increased Participation: The assistant will lower the barrier to entry for potential DReps, encouraging a more diverse range of community members to participate in governance.

Education and Skill Development: As DReps interact with the AI-Assistant, they will gain deeper knowledge of cryptography, blockchain technology, and governance principles, elevating the overall expertise within the community.

Transparency and Trust: By making complex proposals and their implications more accessible, the AI-Assistant will foster greater transparency in the governance process, building trust among stakeholders.

Efficient Resource Allocation: Better-informed decisions will lead to more effective use of community resources, accelerating Cardano's development and adoption.

Innovation Catalyst: The AI-Assistant could serve as a model for other blockchain projects, positioning Cardano as a leader in innovative governance solutions.

Measuring Impact:

• Number of DReps onboarded
• User satisfaction and social visibility (testimonials)
• Amount of text based data with Cardano, Catalyst and liquid democracy content
• Fluency in topics such as cryptograohy, political science, values, ethics, philosophy
• Queries to train the assistant
• Amount of "crude data" measured in data chunks and GB

What is your capability to deliver your project with high levels of trust and accountability? How do you intend to validate if your approach is feasible?

Sapient received Catalyst funding for two proposals in Fund-11 and one of them has been fully completed and already listed among completed proposals (Catalyst Public Reporting Tracker). The other has had Proof of Achievement for 3 Milestones approved and incorporated a demo open source LLM with many Catalyst and Cardano datasets included in its training, including several years of 4 different Telegram channels chat history, all previous voting, proposal, community assessment and moderation information, among others. Experimenting with many open source LLMs, generative AI custom agents, Cardano data, LangChain and Llama 3.1. for the purpose of a secure, specialized ecosystem application put us in an excellent position to provide a highly customized DRep suite of AI assistance and support.

We are experts in wisdom of crowd consensus and human - AI interfaces. Our team members have worked decades in TradFi markets, hold Master of Finance and quantitative finance degrees and work with financial and commodity companies for trading, risk management and decision making software. With backgrounds in many different asset classes and jurisdictions, we bring a broad understanding to make real world asset tokenization work beyond crypto novelty use cases.

Sapient Predictive Analytics has been in the data science, machine learning and collective intelligence space since 2018, winning awards and recognition in Singapore and beyond. We have previously worked with Fraunhofer Institute, Macquarie bank and many small and medium size companies for their decision making, governance, data analytics and trading desks.

Our funded Data-Driven Catalyst system improvements proposal from Fund-11 explores Catalyst-specific chat, for example to give new users a simple and interactive way to learn about complicated topics such as fund rules, timeline, requirements and good proposal drafting, the community review and moderation process, onboarding and milestones, proof of achievement and project close-out.

What are the key milestones you need to achieve in order to complete your project successfully?

Milestone 1: Foundation and Data Gathering (~6 weeks)

Outputs

Compile comprehensive knowledge-base on Cardano, Voltaire, DRep code, node operations, blockchain technology, and governance principles

Develop initial LLM model architecture suitable for most relevant queries and languages

Decision on suitable KPIs for the project (part of SOM)

Acceptance criteria

Clear and thorough discussion and decisions on scope, project management and technology

Most up to date inputs for example CIP-1694 workshops

Evidence

All documentation is publicly available

Code is open source and published on Github

Video walk-through

Milestone 2: Core LLM Development (~8 weeks)

Outputs

Train AI model on the compiled knowledge base

Implement natural language processing for proposal analysis

Develop risk assessment module

Acceptance criteria

Code passes tests and does what is expected

Clear documentation and blueprints

Demo of first interactions, look-and-feel

Evidence

All documentation is publicly available

Code is open source and published on Github

Milestone 3: Production Data Training and User Interface (~6 weeks)

Outputs

Cardano- and DRep-role specific context and interactions

Design and develop user-friendly interface for DReps

Integrate AI model with Cardano governance and tooling

Acceptance criteria

Significantly larger data volume and semi-supervised training process

Clear documentation and project progress

Evidence

Video or demo of feeds and how they are consumed

All documentation is publicly available

Code is open source and published on Github

Milestone 4: Advanced Training and Refinement (~8 weeks)

Outputs

Conduct extensive testing with a select group of DReps

Refine AI responses and user interface based on feedback

Create educational content and technical on-chain integration

Acceptance criteria

Assistant design is clear and suitable for beta cohort requests

Clear documentation and project progress

Evidence

Video or demo of feeds and how they are consumed

All documentation is publicly available

Code is open source and published on Github

Milestone 5: Compliance and Ecosystem Integration (~6 weeks)

Outputs

Optimize the platform for handling a large number of diverse requests

Develop API and offline functionality

Implement understanding of more complex moral, diversity or decentralization topics

Ensure compliance with data protection regulations

Acceptance criteria

Design is clear and complex queries are explained well

Clear documentation and project progress

Evidence

Video or demo of functionality

All documentation is publicly available

Code is open source and published on Github

Final Milestone: Outputs

Close-out report

Close-out video

Full launch and social media visibility

Acceptance criteria

Close-out reports are available publicly

Proof of actual DReps on Cardano using our product

Comprehensive nature of innovative aspects in comparison to existing products and servies

Evidence of milestone completion

Github code and documentation available for all outputs, public video-walkthrough

Who is in the project team and what are their roles?

Thomas Wedler: project management and trading-related content

Experienced financial trader and entrepreneur. Ex Shell, Vattenfall, Masefield senior futures and options trader. Individual floor trader at Singapore Exchange. Tom has been building and deploying programs for automated market making and energy derivatives since 2014. 15 years Derivatives experience at multi-national organizations working closely with industry bodies and speaker at market conferences and workshops. Involved in crypto trading since 2014 and DeFi/oracles since 2018. Plutus Pioneer, Marlowe Pioneer and Atala Prism Pioneer.

Thomas is a certified Superforecaster with the Good Judgment Project and winner of the inaugural Hybrid Forecasting Challenge at SAGE / University of Southern California.

https://www.linkedin.com/in/thomas-wedler-18960/

Role in Catalyst: Challenge Team (Fund 8-10), Sub-circle3, Catalyst Coordinators (funded proposers), Veteran Proposal Assessor, Reviewer in funded project milestone reporting (PoA pilot) Fund 9 - 12.

June Akra: project management and risk-management related content, community

Sapient developer team: to provide UI front-end and API for the portal

Founding member of BlockCarbon, financial market expert and academic with vast experience in risk management, derivatives and commodities. Experience for various risk functions in 2 billion dollar AUM fund. Holder of Master degree in Investment with distinction and awarded Draper Prize. Certified Quantitative Finance (CQF) alumni London. Experienced video editor, content creator with combined 50,000 followers on social media, NFT collector and creator. Certified python AI practitioner, Plutus Pioneer &amp; Atala Prism Pioneer.

https://www.linkedin.com/in/june-a-a3a0b4174

Role in Catalyst: Challenge Team (Fund 7-10), Sub-circle3, Catalyst Coordinators (funded proposers), Veteran Proposal Assessor, Reviewer in funded project milestone reporting (PoA pilot) Fund 9 - 12.

Sapient team members upon demand:

Data scientist, Senior fullstack developer / head architect, LLM engineer, data engineer

https://www.18hall.com/sapient-predictive-analytics/

Please provide a cost breakdown of the proposed work and resources

Foundation and Data Gathering

25% or 104,500 ADA

Technical design and project planning

Comprehensive knowledge-base on Cardano, Voltaire, DRep code, node operations, blockchain technology, and governance principles

Development for initial LLM model architecture suitable for most relevant queries

Estimated hours: 1,200 hours

Core LLM Development

25% or 104,500 ADA

Training of AI model on the compiled knowledge base

Implementation natural language processing for proposal analysis

Development risk assessment module

Estimated hours: 1,200 hours

Production Data Training and User Interface

15% or 62,700 ADA

Cardano- and DRep-role specific context and interactions

Design and develop user-friendly interface for DReps

Integrate AI model with Cardano governance and tooling

Estimated hours: 720 hours

Advanced Training and Refinement

20% or 83,600 ADA

Extensive testing with a beta cohort/group of DReps

Refinement of AI responses and user interface based on feedback

Educational content and technical on-chain integration

Estimated hours: 960 hours

Compliance and Ecosystem Integration

15% or 62,700 ADA

Optimization of the platform for handling a large number of diverse requests

Development of API and offline functionality

Implementation of more complex moral, diversity or decentralization topics

Ensuring compliance with data protection regulations

Estimated hours: 720 hours

Project close-out and community reporting will be provided free of charge.

<u>Ada exchange rate management</u>

Since the last Catalyst fund, Ada to USD has been relatively stable and our budget is based on $0.35 or 0.45 SGD per Ada. Our team are used to fluctuations as we received funding for 2 projects in F-11 when large volatility occurred. We will overdeliver on training data inputs and optional features in a bull market and have sufficient cash flow and resources to fully deliver the project in a bear market.

No dependencies.

How does the cost of the project represent value for money for the Cardano ecosystem?

The DRep AI Assistant can provide outstanding value for the ecosystem and especially treasury by safeguarding Ada available to the governance function. The overall level of knowledge and measured assessment will increase dramatically as a result of a constantly learning and updating assistant LLM and help introduce functions that go beyond convenience. Fairness and contribution to the common good of our ecosystem will benefit from risk mitigation and access to relevant information and data beyond any single DReps capabilities.

Many DReps are likely to be influencers and stake pool operators who may benefit from pointers on cryptography, philosophy, or political science. The AI-Assistant is designed to supplement their knowledge help mitigate various risks:

Knowledge Deficit vs Knowledge

• DReps lacking sufficient knowledge for effective governance.
• Provide on-demand, in-depth explanations of complex topics, tailored to the DRep's current understanding, including CIP 1694 and its governance system. Offer continuous education on blockchain technology, Cardano-specific concepts, and governance principles.

Technical Misunderstandings

• Misinterpretation of technical proposals leading to poor decisions.
• Offer clear, jargon-free explanations of technical concepts and their potential impacts.

Time and Resource Constraints

• DReps unable to thoroughly research all proposals.
• Quickly summarize key points of proposals and provide relevant background information, saving DReps valuable time.

Bias and Influence

• DReps being swayed by personal biases or external influences.
• Present objective, fact-based analyses of proposals and their potential consequences.

Accountability

• Lack of transparency in decision-making processes.
• Log decision-making rationales and provide explanations that DReps can share with their delegates.

Voter Fatigue, Lack of Resources

• Overwhelming number of proposals leading to rushed or skipped votes.
• Prioritize and categorize proposals, highlighting the most critical issues requiring attention.
• DReps with insufficient resources (time, incentives) to properly conduct votes.
• Streamline the voting process by summarizing proposals, highlighting key points, and providing quick access to relevant information, thus maximizing the efficiency of DReps' limited time and resources.

Anti-Sybil Attack Measures

• Widespread sybil behavior, regulation risks in geographical regions, lack of accountability.
• Assist in verifying and maintaining DRep authenticity by providing tools for transparent record-keeping and decision documentation. Offer region-specific regulatory information and help DReps maintain compliance.

Organization Disclosures

• Lack of transparency in DRep affiliations and potential conflicts of interest.
• Guide DReps in creating comprehensive, standardized disclosure statements. Maintain an up-to-date database of organizational affiliations and help identify potential conflicts of interest in voting situations.

Voting Decision Rules

• Inconsistent voting behavior, potential for random voting or botting for incentives.
• Help DReps establish and maintain consistent voting criteria. Provide analysis of voting patterns to ensure alignment with stated principles and detect anomalies that might indicate botting or random voting.

Interests of the Proposer

• Potential conflicts of interest, game theory risks such as vote selling.
• Analyze proposals for potential conflicts of interest. Provide game theory insights to help DReps understand the long-term implications of their voting decisions.

Coalition Measures

• Powerful voter collusions between DReps, potential for an overpowering "superstar" DRep.
• Monitor voting patterns to detect potential collusions. Provide insights on voting power distribution and alert DReps to situations where power may be overly concentrated.