DreamMaker AI - the future of human-machine interaction

Imagine a companion always by your side, understanding your deepest needs and aspirations. More than just an assistant; empowering you to achieve your dreams while freeing you to truly live.

Comment

Inspiration

Imagine a companion always by your side, understanding your deepest needs and aspirations. DreamMaker is more than just an assistant; it's your virtual life partner, empowering you to achieve your dreams while freeing you to truly live. This groundbreaking tool utilizes cutting-edge AI technologies to understand human needs, goals, and even emotional states. DreamMaker acts as a tireless partner in the digital world, proactively working in the background to support your aspirations.

What it does

DreamMaker acts as your personal AI coach, confidante, and guide. It goes beyond the limitations of traditional Q&A systems, offering a dynamic and multifaceted experience. Currently, we implement the Financial Planning feature: DreamMaker analyze your financial situation and creates a personalized plan to reach your financial goals. It provides insights into your spending habits, helps you find new income opportunities, and ensures you're on track for financial stability.

How we built it

  • Design: From our business objective, we designed our LLM and data system with the idea of modularization in mind. Most components are isolated. Not only to support the change in business opportunity, but it also embrace technology evolution, especially in this LLM field as it is still very active in research and development.

  • LLM: We have tried deploy Dolly using Sagemaker. It requires some workaround to use. However, the actual challenge of our work is LLM context length. Dolly allows us to use up to …. token whereas OpenAI could handle twice as much. To cope with Dolly, our LLM architecture has to be more recursive and dynamic to overcome max token size per request.

  • We use Langchain framework to put together LLM architecture and process. It accelerates our development by providing somewhat pre-defined structure of the way to interact with LLM. We also extend some tools with python to support our business objective.

  • Vector database: we have used ChromaDB to setup our prototype. In the future, this part change be changed to others product. This area of work required a proper data model design organize and manage document processing work.

  • Business Data: We only used CSV file to get the impression of how our e2e process look like. We actually planned to store relational data in Delta table, which is suitable to manage data lifecycles in business environment as they tend to be more structured.

Challenges we ran into

  • response time
  • qa used to much token (cannot use too much data)
  • memory cannot carry over for long due to token limit
  • incorrect tool interpretation
  • unexpected behaviour (response structure or wording for LLM) so the parser cannot perform properly
  • a prompt cannot be generic, we need to build agent or chain manager to orchestrate the flow
  • require manual Dolly serving
  • require adapter to convert information between dolly and langchain

Accomplishments that we're proud of

Despite the challenges, we've achieved significant milestones:

  • LLM Integration: We successfully integrated LLMs into our solution, demonstrating their potential for business applications.
  • Business-Specific Design: We tailored DreamMaker to effectively address specific business needs.
  • Continuous Improvement: We envision using LLMs as part of the solution to our business objective, though there're challenges to overcome.

What we learned

The journey to create DreamMaker has been a valuable learning experience:

  • LLM Implementation: Integrating LLMs effectively requires careful planning and execution in the business environment. We need to work around prompt engineering to construct the most relevant information, which, sometimes, can be huge and exceed token size.
  • Industry-Specific Challenges: Existing industries have unique challenges when utilizing LLM technology.
  • Prompt Engineering: Crafting precise prompts is crucial for optimal LLM performance.
  • Word Sensitivity: Word choices for input, prompt, and response significantly impact LLM behavior, requiring careful consideration. LLM can misbehave with different words.

What's next for DreamMaker AI

DreamMaker's potential is limitless. We're committed to adding exciting new features:

  • Marketplace: Connect with partners, find products and services, and access professional support.
  • Deeper Connections: Foster a more meaningful relationship with your AI companion through initiative and personalized interactions.
  • Community Building: Connect and share experiences with others who share your values and goals. Life Progress Tracking: Monitor your progress towards your goals and receive personalized recommendations.
  • Gamified Incentives: Stay engaged and motivated through interactive features and rewards. With each advancement, DreamMaker will become an even more integral part of your life, empowering you to achieve your dreams and live a fulfilling life.

Built With

Share this project:

Updates