Harmonica — superfast deliberation using GPT-4

Harmonica is a tool that enables fast and scalable sensemaking & deliberation by talking to org members 1:1 and using LLM to synthesize their responses and ask additional questions if needed. It saves a lot of time, makes it easy to follow collective intelligence protocols, and increases participation in governance.

The main objectives for Harmonica at this stage:

• Build a chatbot solution that synthesizes natural language input from multiple users for collective decision-making. We will complete this goal by launching the MVP within 2 months of funding.
• Make DAO governance more intuitive and engaging. We will measure progress toward this goal through case studies with early adopter DAOs showing decreased time to reach decisions and increased participation rates after installing Harmonica.

The Problem

Organisations (especially larger ones) face serious challenges with employee participation in decision making that negatively affects their engagement. Thoughtful deliberation doesn't scale. People feel voiceless and disrespected. Valuable insights are missed.

On the other hand, democratically governed orgs (e.g. employee-owned businesses, OSS, DAOs) choose between conflict-prone and slow open decision-making and untrusted closed decision-making, which puts their community health at risk.

The Solution

Generative AI + conversational UX = chatbot for superfast deliberation / sensemaking. By talking to participants 1:1 and synthesizing their responses, our chatbot enables more people to participate in important discussions and saves a lot of time.

Harmonica aims to improve collective decision-making by using GPT-4, Discord chatbot and facilitation templates to synthesise participants' input into consensus points or original solutions without them talking to each other directly. Harmonica removes conflict from the deliberation process and increases engagement, enabling orgs to coordinate efficiently. Participants can be candid with their opinions without offending others, hear positions and arguments independent of individual personalities, bypass organisational politics, and still benefit from a deliberative process that bridges opinions and supports alignment.

Roadmap / milestones

1. Proof of Concept [completed] ✅

Build a proof of concept to demo with potential users and experts in the field of governance, validate the concept and get feedback on best-practice methods for building and structuring an LLM-powered deliberation tool.

• Build POC Discord bot with basic template ✅
• Consult with practitioners and experts ✅
• Outline initial roadmap ✅

2. MVP (end of 2023)

The goal of the first milestone is to build the MVP. It should offer a basic facilitation experience throughout the whole decision-making journey, and have sufficient functionality to validate the demand in the pilot study.

• Discord bot with improved UX
• Web dashboard
• 3+ useful facilitation templates (prompt engineering)
• Separate dialogues with session participants
• Product analytics

3. First revenue (Q1 2024)

The goal of the second milestone…

• Payments (introduce subscription / pro features)
• 10+ facilitation templates + an easy way to add them
• Slack + WhatsApp support
• Improvements of web dashboard
• Activation of customer’s LLM licence key

4. Advanced sensemaking (Q2 2024)

The goal of the second milestone…

• Embedding additional context (responses from previous sessions, handbooks and constitutions, etc.)
• Using Langchain and DBs to process more info / ask better questions
• 30+ facilitation templates + SDK
• Templates store
• Proprietary chat on web
• Harmonica API and documentation

Team

Artem Zhiganov — Product strategist with background in marketing and user research. Former core member of Protein Community. Governance nerd. University of Manchester alum (MSc Innovation Management), Newspeak House fellow.

Harry Eastham — Brand & Interaction Designer leveraging AI, data & no-code tools to produce (almost) anything. Built & launched Gitcoin's new website & brand. Previously founding team of Protein Community, design lead at Here Be Dragons.

James Eastham — Senior Cloud Architect at Amazon Web Services and content creator. He has over 10 years experience in software, at all layers of the application stack.

Joshua Becker — Assistant professor at UCL School of Management, PhD researcher of group decision making and collective intelligence. Joshua brings his immensely valuable expertise in facilitation and informs our experiments in terms of heuristics / prompt engineering.

Will Payne — Algorithms engineer, systems engineer and technical lead with over 20 years of experience developing ML/AI intensive systems. Most of this experience has been in the aerospace, maritime and automotive industries, and has revolved around the development of ML/AI intensive sensing systems that have safety and security requirements.

Andrea Gallagher — Researcher with 25 years experience in information architecture, interaction design, and UX strategy. Started in the earliest web1 startups, through dot.com consultancy, then focusing on User Research and Design Management at Intuit, Google, and Asana. In the last few years she’s been active in web3, leading User Research at Aragon and now at RnDAO. Trained and taught design thinking and group facilitation, including work as an Innovation Catalyst at Intuit.

Early prototype: https://www.loom.com/share/d9c165142d62448cb2e0895138e39a52

Test server with the latest version: https://discord.gg/MWr8dnKJ37