Qiongting Zhang

The Accelerator


Local Projects
NLP / Python / GPT / Twilio / AWS / MongoDB / OSC
Aug.-Oct. 2023, New York.


Role:
Chatbot Developer, Back-End Developer

Collaborator:

Jake Barton, Ai Xia, Tan Xiao, Ker Chen, Michael Edgcumb, William Hardy, et al.


OVERVIEW


"The Accelerator" is an immersive "Time Machine" experience that transports participants into the future, allowing them to interact with their future selves and take action to shape a more sustainable world.



Within the "Time Machine" (a Dreamcube—a room composed of four screens), players converse with a future-me chatbot via text messaging, which will prompt participants to contemplate carbon emissions issues and consider actionable steps they can take in their daily lives, leveraging their current positions and fields of work.

Partial Future-Me Chatbot Interaction Schematic

As they engage with the chatbot, participants witness dynamic environmental changes unfolding around them, with the ability to interact with their surroundings.

Demonstrations of Interactions within the Dreamcube

By making carbon emissions and individual actions visible, "The Accelerator" can help you discover one single action that will make a difference.

Where will your journey start?


Future-Me Chatbot


To facilitate dynamic conversations with future selves, I integrated the OpenAI API for natural-sounding responses. Ensuring directed dialogues, I combined NLTK and Spacy capabilities seamlessly with the OpenAI API throughout.

The chatbot interaction process is divided into four distinct parts:

1. Introduction and Information Gathering


The chatbot initiates the conversation by introducing itself and collects relevant information from the user, including name, age, expectations for future. This phase sets the stage for personalized interactions based on the user's background.

  
Guided GPT Conversations with NLTK and Spacy Monitoring
Information Collection by GPT


2. Guided Solution Selection


The chatbot guides users through a series of choices, prompting them to select specific solutions and techniques that align with their fields of work or daily activities to reduce carbon emissions. Encouragement to collaborate with others and extend invitations to join these actions is also provided. Once the selection process is complete, users are transitioned into the "Time Machine."

  
GPT-Prompted Responses and Fuzzy Matching with NLTK


3. Time Machine Experience


Within the "Time Machine," users visually witness their selected actions unfolding on screen, complemented by a narrative crafted by ChatGPT. This story details the outcomes that follow their initial actions, using information gathered during the chat dialogue. As the narrative unfolds, the surrounding environment progressively improves, illustrating the positive impact of their chosen actions.


4. Follow-up and Encouragement


Upon completing the experience, users receive the story via message. They can continue to engage with the chatbot in subsequent conversations, receiving encouragement and reminders about environmentally friendly actions at opportune moments.

  
GPT Story Generation from Collected Conversation Data
Freeform follow-up Chat with GPT



The chatbot's texting function, using Python via Twilio, is hosted on AWS, with user data automatically saved in MongoDB. Story and other information are transferred to Unreal Engine using VPN and OSC.