Inspiration
Being stuck in traffic is a mind-numbing, time-consuming, environmentally detrimental, yet universal experience in today's ever-expanding urban world. A significant portion of energy use in the U.S. is solely for transport. This consumption is a contributor to air pollution and carbon and nitrogen emissions. We hope to optimize public transport routes in order to reduce traffic congestion and working towards the global initiative of reducing carbon emissions to combat climate change.
What it does
Choosing a section of a map shows a traffic density heatmap of the area. This is passed to Claude 3.5, which provides a street view of exact latitude and longitude coordinates for locations to add bus stops, as well as the reasoning behind choosing these locations to lower overall congestion.
How we built it
We used React frameworks to build the front end. The satellite map was created using Here APIs. We created a heuristic using multiple data points to provide a proprietary traffic score. We used this score to create a gradient map to signify areas with higher traffic emissions. We prompt engineered Claude 3.5 through systematic analysis, weighing the pros and cons of each proposed solution before making a final recommendation. We selected Claude 3.5 Sonnet because of its 200k token context window. It also has superior computer vision capabilities in chart interpretation and OCR compared to opensource LlaVA, which are crucial for our geospatial traffic analysis. The back end relied on Django frameworks.
Challenges we ran into
There was a learning curve when integrating the Bedrock API, as it was the first time we used this technology. We overcame this by poring over documentation and using different integration approaches. Our proprietary traffic scoring algorithm relied on balancing multiple data points. This posed a technical challenge as well.
Accomplishments that we're proud of
We are proud of the fact that we are able to produce an accurate current heatmap of traffic and use it to generate a traffic score. Our outputs from Claude have sound logical reasoning and potentially implementable solutions.
What we learned
We learned a great deal throughout our implementation of Urban Alchemist. As we stated, we learned a lot about implementing Bedrock APIs. We gained insight on factors that affect urban transportation systems. Most importantly we learned about the potential of AI to provide solutions to mitigate the effects of climate change.
What's next for Urban Alchemist
We plan on implementing a transit layer and a bicycling layer from google maps APIs to add to our map. We also plan on analyzing urban space to include suggestions on where to add solar panels as a form of alternative energy.
Log in or sign up for Devpost to join the conversation.