VanData

Main Page
Traffic Map
Pollution Map
Chatbot
Chatbot
Batch Video Processing with Modal Video Model
Requests, tagged by Modal Image Model
Request Form
Editable list of Cameras and Standard Video Processing
Accessible to All Languages
Sign In
Profile
Top Bar

Inspiration

Traffic in Chicago is horrible, with three main airports, the Bean, and the most beloved pizza in the Midwest, its no wonder why the streets are always live. But how much of this live traffic is being recorded? How much pollution is being emitted to the environment? And how is that affecting the communities of this diverse city? As two people who have witnessed the congestion, and two people who have lived their whole lives within such streets, we decided that we wanted to make a project to educate and make a change for the Windy City we all love.

What it does

VanData's main purpose is to inform, represent, and act. With the resources of the City, the computational power of Modal/Snowflake, and community input, VanData fights vehicle pollution with AI.

How we built it

Our project is divided into three parts: the vehicle detection model, the website, and the Snowflake RAG Chatbot. The traffic video data is fed to the model and processed to output the vehicle types and count. Then, we analyze the data on Snowflake to feed to the website, which is primarily used in our traffic/emissions heatmap. Snowflake also serves as the RAG database for the chatbot on the website.

Challenges we ran into

Model Navigation We've spent some time searching for a viable model to identify vehicles. This was core to our project as both accuracy and speed were required. We labeled various testing data and tested out different models and ultimately landed upon one that perfectly fitted our needs.
Data Processing We've debated on numerous ways of getting data. Like, "where would we get our data?", "what form and length should the data be?", and "should we allow all users to provide data?" Eventually, we went with having the app being contracted out to local city government, as they have the resources and authority to set up 24/7 cameras. We understand that budgets are tight, which is why VanData can work with practically any modern webcam, as opposed to the $60,000 cost of installing just one traffic camera. To supplement the data, residents are allowed to take pictures of roads and submit reports of real-time traffic situations. With enough reports, city government will be encouraged to respond to the demand and install cameras to monitor the traffic in underserved communities.

Accomplishments that we're proud of

We are honestly very proud that we got this far at all, for half of the team this was their first hackathon ever, and this was a project that involved a lot of moving parts. Our favourite features are the Snowflake Chatbot, which we have provided with a wide range of reputable vehicle emissions information as well as our entire database, and the batch upload feature, which takes full advantage of Modal's multi-container capabilities.

What we learned

We learned how to develop database schemas and be able to converse about them as a team, as the database was an important part in everyone's share of work. We also learned about the strengths and weaknesses of different LLMs, and how to chunk document data to improve RAG performance.

What's next for VanData

We would love the opportunity to discuss this project with the city officials in Chicago. It would be a dream come true for our team to see this implemented in our very own hometown.