Tryp

Inspiration

Frustrated with traffic given the increased population growth and tourism. Other countries (for instance in Europe) have solved the problem of mass transit. It seems to be a American culture issue (in particular in the southern states). What if we can use AI to motivate people to do the right thing? Maybe we just need to pay them money to do the right thing. Economic incentives to change behavior patterns and establish new routines using AI, which can benefit humanity as a whole.

What it does

-Use AI Models: Utilize AI to analyze real-time and historical traffic data.

-Custom Notifications: Send push notifications suggesting optimal travel times and routes.

-Example: “Based on current traffic, the 8:15 AM bus from Summerville to downtown Charleston will save you 20 minutes compared to driving. Catch the bus and get a $2 discount at Joe’s Cafe!”

“Join 500+ downtown commuters who are taking the bus today to avoid the I-26 traffic jam. Enjoy a FREE BEER with your bus ticket!”

How we built it

Our app leverages data from TomTom's Traffic Incidents and Traffic Flow APIs to provide real-time insights into current traffic conditions. This data is sent to the OpenAI API's language model, which processes and analyzes it to predict traffic density, estimated time to destination, and potential cost savings for each trip. By combining these advanced technologies, our app can offer users accurate and actionable travel recommendations, helping them make informed decisions and save money on their daily commutes.

The AI is wrapped in a simple Flask application that the frontend calls for its passenger predictions.

The AI prediction model is using the amazon/chronos-t5-mini model for "pretrained time series forecasting based on language model architectures". It enables "zero-shot" prediction -- in other words, the model is pre-trained across multiple domains of time-series data and can make predictions off of any time-series data without having been explicitly trained on it. The results, even using this smaller version of the model, are impressive. The data used is from a DATA.gov public dataset of NY MTA bus ridership data across multiple years.

We used the AI model to predict, to either a future day or future hour, the amount of passengers that may be riding the bus.

Frontend: React

Next.js bootstrapped with create-next-app

ChatGPT's API

Prompt to evaluate traffic conditions and necessary financial incentives

Backend:

PyTorch

Prediction model

Ingest historical traffic data and bus ridership to suggest time/play to push economic incentive

Challenges we ran into

Can we find enough suitable data for free to feed into AI. Which kind of financial incentives can change behavior and persuade people to ride the bus when they typically "don't want to" (a human psychology challenge)

Accomplishments that we're proud of

Collaboration between team - we didn't know each other beforehand and met at the hackathon Prompts - used ChatGPT to improve app, determine financial incentives and user scenarios, as well as improve prompt based upon our feedback

What we learned

Many people hate traffic, and wish buses were a more used option here. Also, the power of AI to help us flesh out an idea and build a prototype from scratch

What's next for Tryp

AWS - Serverless function - Lambda with S3 bucket and Bedrock/Sagemaker Integrations with existing transit apps (like "Transit" app that CARTA uses for bus trip planning)

Built With

• openai
• pytoch
• react
• typescript