Inspiration

The definition of hell is that on your last day on earth, you meet the person you could have become. This rather visceral reminder that we are not living up to our potential is backed up by statistics; studies suggest that 92% of us fail to reach our goals. Setting goals is easy, achieving them is hard.

What makes the aforementioned definition of hell so gut-wrenching is you only meet your best self when your life is already over and it’s too late to do anything about it. Imagine how powerful it would be to meet your future best-self today.

Studies show that visualisations are a powerful tool for achieving our goals. However, effective visualisation is a skill in itself. With AI technology, it is possible to build a tool that makes goal visualisation accessible for everyone.

A personal note on the inspiration behind this project: I (Sarah White), am a highschool dropout turned serial edtech entrepreneur. I know, perhaps more than most, the importance of education, personal and professional development in unlocking opportunities for a better life. Helping people reach their full potential, especially those who did not benefit from a traditional mainstream education, is a life-long obsession. Tools like newyou.ai are a great way to bring focus and discipline to people who want to level up but need extra help to do so.

What it does

Put simply, newyou.ai enables you to talk with your future self. Your future self encourages you to work towards your goals and even gives you actionable steps to help you. Having your future self describe the feeling of achieving your goal, and what impact it has had on your life, is a visceral reminder of your future potential.

A step by step flow: newyou.ai asks you for several pieces of personal information; your name, what goal you would like to achieve, when you would like to achieve it by, why you want to achieve this goal, and how it will make you feel. Next, newyou.ai uses it’s time machine (aka ChatGPT4) to create a your future persona. Your future persona then messages you, letting you know how great it is to have achieved your goals, some of the struggles you encountered along the way, and thanking you for your hard work to make it possible. Your future persona also sends you a list of actionable steps to take to work towards your goal. Plus, your future self will drop you a message from time to time to motivate you on your journey.

How we built it

Voiceflow Platform Setup: We began by creating a new project within the Voiceflow platform, defining the basic conversational structure and user intents that would drive the chatbot's interactions.

Conversation Flow Design: We built a complex conversational flow using Voiceflow's drag-and-drop interface, incorporating components such as intents, variables, conditions, and other logic elements to create dynamic and context-aware interactions.

Integrating ChatGPT-4 API: Within Voiceflow, we utilized the pre-integrated ChatGPT-4 API to handle natural language understanding and response generation. By configuring the settings, we were able to generate relevant and engaging responses based on user inputs.

WhatsApp Integration: We connected the chatbot to WhatsApp using Voiceflow's built-in WhatsApp integration feature. This involved ensuring the proper parsing of user inputs, and managing conversation states for maintaining context across interactions.

Error Handling: We implemented error handling mechanisms within the conversation flow to address unexpected user inputs and edge cases. This included the use of fallback intents, context-aware redirection, and providing helpful responses even when users went off-script.

Testing and Debugging: We conducted extensive testing and debugging of the chatbot within Voiceflow's platform and through WhatsApp, simulating a wide variety of user interactions and edge cases. We iterated on the conversation flow based on test results to improve its performance and reliability.

Performance Optimization: We optimized the chatbot's performance by minimizing API calls, efficiently using variables, and refining conversation logic within Voiceflow. This ensured that the chatbot provided quick and accurate responses, enhancing the overall user experience.

Challenges we ran into

Scope change: Initially, we were going to build something quite different. The original idea had the same inspiration, but the output was going to be visual, not text. The user would input the same information as they do now, however, they would also input several photos of themselves. Newyou.ai would then use AI to render images of the user achieving their goal. (for example, if my goal was to run a marathon, I’d see an image of myself running a marathon). We decided that in the time frame available, a chat based option was a more effective solution.

Model Optimization: One of the main challenges was optimizing the GPT-4 model to provide real-time, contextually relevant responses. We addressed this by using techniques such as hyperparameter tuning, model pruning, and incorporating a custom tokenization strategy to reduce latency and improve response quality.

Conversational Context Management: Maintaining the context of ongoing conversations within the Voiceflow Dialogue Flow and managing user expectations proved challenging. We addressed this by mapping user intents and using conditional logic, variables, and context-based responses to create dynamic conversations.

Managing User Expectations: Ensuring that the chatbot met user expectations in terms of conversation quality and relevance required careful design of intents and responses. We addressed this by refining the training data used for the GPT-4 model and iteratively improving the chatbot's responses based on user feedback.

Handling Errors and Edge Cases: Ensuring that the chatbot could gracefully handle unexpected inputs and edge cases was crucial. We addressed this by incorporating error handling mechanisms, fallback intents, and appropriate redirection within the conversation flow to provide helpful responses even when users provided unexpected input.

Testing and Debugging: Conducting thorough testing and debugging of the chatbot within Voiceflow's platform and through WhatsApp was essential to ensure its reliability. We addressed this by simulating various user interactions, testing edge cases, and iterating on the conversation flow to improve its performance.

Performance Optimization: Ensuring that the chatbot provided quick and accurate responses required optimization of the conversation flow within Voiceflow. We addressed this by optimizing the usage of variables, and refining the conversation logic to reduce latency and improve the overall user experience.

General: And of course, just the general challenge of coming up with a concept and delivering it within such a tight time frame. This challenge was quickly overcome by our shared vision and ability to communicate effectively.

Accomplishments that we're proud of

Implementing a robust NLP pipeline using the GPT-4 model for generating personalized goal-setting advice and motivation, which involved fine-tuning the model on a custom dataset and optimizing it for real-time interactions with users.

Developing a highly responsive and adaptive conversation flow using a state management system that tracks user inputs and adjusts the dialog based on their needs, employing techniques such as sentiment analysis.

Incorporating modern design principles and accessibility features to ensure a positive user experience.

Pivoting the platform from an image to a text solution 18 hours before the deadline was a risk that not only we are glad we took, but are proud to have delivered a solution that delivers a great service.

We believe that newyou.ai could be extremely beneficial to many people. However, even if we just helped one person achieve one goal that was important to them, that alone would be an accomplishment to be proud of.

Also, we are both new to the hackathon world, so we are both proud to have taken the leap and participated in this competition!

What we learned

Advanced techniques for fine-tuning and optimizing large language models like GPT-4 to generate more accurate and contextually relevant responses.

Leveraging modern platforms to build a scalable and high-performance application.

Applying best practices for optimizing query performance to ensure smooth user experiences.

What's next for newyou.ai

There is so much potential for newyou.ai, and we’re excited to explore various avenues of development;

Features - images Build the visual part of the platform, as mentioned in the ‘challenges’ section. Images will be a powerful addition to the 'visualisation' aspect of newyou.ai

Features - voice Include voice notes, delivered in the user’s voice. Hearing your own voice describe your future life will increase your emotional investment in the goals you are setting

Community Build a community of people who want to work on their professional development, who will serve as our early adopters / testbed for new features

Additional resources Deliver additional resources to our (future) community to help support them in their personal and professional development journey, including podcasts, YouTube channel, blogs etc

Research Conducting academically robust research into the efficacy of newyou.ai for achieving goals would be an exciting step in validating our unique approach to setting and achieving goals

Building a real time machine The ultimate version of newyou.ai would be a real time machine. We’re just waiting for technology to catch up with our vision.

Built With

  • chatgpt-4-llm
  • voiceflow-platform
  • whatsapp
Share this project:

Updates