KOL [The Girlfriend Retainer]

Inspiration

The inspiration for KOL [The Girlfriend Retainer] came from a deeply personal place. Each of us on the team has experienced the pain of miscommunication in relationships, whether with a partner, family member, or close friend. We've each felt debilitating heartbreak that impacted everything from our health and studies to career. During our research, we discovered that 63% of relationships fail due to communication issues, it's one of the primary relationship wreckers — stemming from issues that are often avoidable with the right tools. This realization hit home when one of our team members shared a story about a close friend who went through a heartbreaking breakup because of unresolved communication breakdowns. This person was an international student studying in Canada who until then was getting straight As but suddenly lost all motivation to go to class and couldn't concentrate, waitlists for university counsellors were over a year long and this student ended up failing all their classes that term, eventually getting kicked out of university and losing their study permit. This set their entire life trajectory off course. The emotional toll, the confusion, and the regret were profound. What if technology could help us communicate better, before things fall apart?

We realized there was an opportunity to create something meaningful—a tool that could proactively help people navigate the complexities of human connection and preserve the relationships that matter most.

What it does

KOL [The Girlfriend Retainer] is an AI-powered application that transcribes and analyzes conversations in real-time to provide actionable feedback on communication patterns condusive to stronger relationships, identifying triggers and understanding how one's own communication style harms their relationship. By recognizing triggers, tone shifts, and recurring conflicts, the app helps users understand how their words and actions affect their relationships. It offers personalized suggestions for improvement, tracks progress over time, and even uses predictive analytics to warn users when their communication patterns might lead to conflict. There are both positive and negative reinforcement mechanisms to keep users on the path towards continuous improvement in their relationship. Our goal is to help people break maladaptive communication patterns and foster healthier, happier relationships.

How we built it

We built KOL using a multi-layered composite architecture that integrates multiple technologies:

OpenAI's API powers our NLP capabilities for summarizing conversations, categorizing and generating detailed feedback tailored to individual communication styles. -Real-time transcript processors enable the app to analyze tone, sentiment, and intent during live conversations.
Omi's memory creation triggers identify pivotal moments in conversations and store them as retrievable data points for pattern analysis.
Our MVP was developed using a local-first architecture for privacy, ensuring all sensitive data is securely stored and processed locally.
We incorporated predictive analytics to recognize potential conflicts based on historical data and notify users in real-time.
Investigating hardware integration such as using LED light indicators to provide subtle, real-time feedback based on detected stressors.
To further influence behavior, we combined positive reinforcement and optional negative reinforcement:
Positive Reinforcement: Users are rewarded for improvements in communication through visualizations of their progress, celebratory messages, and gamified achievements. These encourage continued use of the app and promote intrinsic motivation to strengthen relationships.
Negative Reinforcement via Lossless Lottery Staking: While not yet live, we’ve deployed a lossless lottery staking contract on Polygon to introduce an innovative element to the app. Users can optionally stake funds to access premium features. If their communication patterns improve, they receive rewards from the staking pool. If they fail to meet improvement benchmarks, their funds are redistributed among other users and the platform, creating a subtle, financially driven incentive to stay committed to their progress. This approach integrates psychology’s commitment devices—methods of reinforcing desired behaviors by introducing stakes (both figuratively and literally).

Challenges we ran into

One of the biggest challenges was balancing privacy with functionality. Analyzing intimate conversations requires robust safeguards to ensure data remains secure and private. Designing a local-first processing pipeline took significant effort especially when integrating with 3rd party APIs like openAPI.

We also faced technical hurdles in achieving real-time processing with high accuracy. Building a system that could understand nuance, such as sarcasm or subtle tone shifts, proved to be a complex task. Omi's Real-Time transcript processors are currently not stable.

On the hardware side, integrating predictive analytics with physical indicators like LED lights was challenging due to the limited customization options of existing hardware. We contacted the Dabl team in EthBangkok and were informed that the Omi devices were meant to be fully opensource and configurable but changing the color of the LED was something that might be possible at a latter point.

Lastly, conducting customer discovery interviews while building the MVP in such a short timeframe was both physically and emotionally demanding. Hearing real stories of heartbreak added urgency to our work but also weighed on us personally. We took advantage of Devcon to interview many people in a short time frame. The idea we had was that we want this to be an actual startup and for that we need to validate our hypothesis, talk to our actual end users before commiting significant development for early signs of product market fit, gauge demand, do market research and ensure there is a sustainable business model present. We found that a freemium or lossless lottery staking model tied in web3 nicely but also resonated most with prospective users.

Accomplishments that we're proud of

Successfully creating a working MVP that integrates real-time transcription, memory triggers, and predictive feedback.
Conducting nearly two dozen customer discovery interviews, which validated the need for our solution and informed its design.
Building a system that prioritizes privacy by keeping sensitive conversations locally stored. Incorporating OpenAI's API to enhance feedback and summaries, making the app feel both intelligent and empathetic.
Developing a roadmap that includes hardware integration and advanced analytics, pushing the boundaries of what AI-powered relationship tools can do.
Taking a startup mindset and approach to maximize chances that this won't be a dead project at the end of the hackathon and will actually be used by real people.

What we learned

We learned that the intersection of technology and human relationships is both promising and deeply personal. Effective solutions require empathy, technical precision, and a commitment to ethical considerations like privacy and data security. AI Agents are blowing up and Omi offers more than a 'Read AI for IRL', we can potentially push forward an entirely new industry of counselling support driven by personalized AI agents that follow couples as they traverse through life's complexities together.

We also discovered the importance of iterative design and customer feedback. Hearing real stories of pain and loss reminded us why we started this project and reinforced our belief that even small changes in communication can have a profound impact.

What's next for KOL [The Girlfriend Retainer]

Enhancing our real-time transcription and predictive feedback capabilities to better handle complex conversational nuances.
Partnering with relationship therapists and communication experts to refine our algorithms and ensure their effectiveness. We want the product to be driven by science and the latest in psychology.
Expanding our hardware integration, allowing users to receive subtle, real-time cues like LED light changes when entering "danger zones."
Launching a beta version with a freemium subscription model to gather more user feedback and fine-tune the experience.
Exploring blockchain-based staking mechanisms to incentivize long-term use and reward users for maintaining healthy communication habits.
Create social channels and start promoting the app.