Reel Rewind - Let's try your story.
Rewrite the movies you love❤️- Amazon PartyRock Product.
Published Mar 7, 2024
I believe the most important reason people enjoy watching movies is because they get to escape into a world of possibilities, experiencing what they cannot in their own lives. Since they can't control their own reality (mostly, of course 😉), While this has always been the case and produced incredible stories, This experience can be enhanced further.
The cinema industry's current limitation is that the audience's imagination and experience are bound by the vision of the movie's directors and writers. To mitigate this, people often discuss movies with their friends and colleagues, or seek out online communities for the same purpose. Countless social media platforms exist where popular movies are publicly and enthusiastically discussed by millions worldwide. As a superhero genre fan, I frequently visit places like https://www.youtube.com/@MissedThePart, countless Reddit communities, and many other public forums to exchange ideas about the movies I like.
This inspired me to create a platform where users could reimagine their favorite movies as they like, giving them personalized experiences. It also features a chatbot with which users can discuss their imaginative scenes and movies, making the experience more interactive.
Reel Rewinds is precisely an experiment to explore the possibility of creating more personalized entertainment for people worldwide.
Here is the project demo: https://youtu.be/DTC5TJgYNVM
I've been experimenting with AI-powered applications for some time now, and I have to acknowledge that AWS PartyRock offers smoothest possible development process for powerful AI applications, providing a positive developer experience.
While developing ReelRewind on PartyRock, the process was fairly straightforward because PartyRock itself manages the interactions between different LLMs, allowing developers to concentrate on the product itself. However, a few improvements could make it even better:
- Expanding model selection: Currently, the choice of models is limited. Providing a wider array of models would enhance developer flexibility.
- Incorporating function calling for data interactions: Enabling LLMs to interact with internet to fetch the relevant information on demand.
I believe addressing these areas would give developers much-needed flexibility and help establish PartyRock as a leading product in this space.
If PartyRock wasn't available, here's how I would use Amazon Bedrock. I would also have considered the factor that due to the nature of the applications, Reel Rewind can highly benefit from being connected to the internet to fetch relevant information at any given moment.
My architecture would consist of:
- Client: Nextjs
- Backend: FastAPI
- Framework for LLM interactions and internet access: Langchain, Langchain tools.
Model Selection
Although, model selection would require experimentation, but as the application is focused on creative writing, I would start my experimentation with the following models:
- Claude 3 Opus (Anthropic): For writing alternate scenes.
- Amazon Titan: For summarizing movie scripts and for conversational chats.
- OpenAI API (Can be called using langchain tools): To generate images using DALL'E 3.
Rough architecture sketch:
The sudden boom in the generative AI space over the last few years has given birth to countless use cases, that range from improving user experiences to creating entirely new ones. Reel Rewind is just one experiment aimed at providing more personalized entertainment to the users around the world. I look forward to building more such innovative and useful applications using AWS.