Google Summer of Code 2023 with the Mayor's Office of New Urban Mechanics: Guidance + Ideas
Below, you will find a list of ideas we have for a Google Summer of Code contributors for the summer of 2023. Thank you for your consideration!
You can find guidance for applications and your project proposals here.
Over the years, we have experimented with chatbots to improve access to City services. Chat via text message or on a website offers a familiar interface for people who might have trouble finding information about a specific program on a website or a PDF.
We are excited by the recent developments of large-language models, and would like to create prototypes with the technology - in particular GPT-3/GPT-4.
This project will include the following activities: 1) Using LlamaIndex (previously known at GPT Index), extending Chat-GPT with external data, such as a PDF about affordable housing services 2) Creating a web app with an intuitive user interface to interact the chat bot. The user should be able to evaluate each response from the chatbot (a simple thumbs up or thumbs down, which should be recorded in a datastore, such as an Airtable base). 3) Evaluating the implementation differences between the API-accessible GPT-3 and Hugging Face Transformers library (for example).
We give this project a medium level of difficulty. The project can be completed in 175 hours.
This project will require intermediate experience with Python, an interest in large-language models and machine learning, interacting with RESTful APIs such as the GPT-3 API, and experience with JavaScript, HTML, and CSS.
The mentors for the project will be a Program Director at the Mayor’s Office of Urban Mechanics, including one who served as a Google Summer of Code mentor at Code for America in 2011 and the City of Boston in 2021 and 2022.
At the Mayor’s Office of New Urban Mechanics, we are focused on helping City departments create innovative services that are accessible, equitable, and delightful. We would like to build a tool that helps City departments keep track of infrastructure investments, as they strive to make them more equitable.
Throughout Boston, City government repairs broken sidewalks, installs benches, builds new libraries, and creates new parks. These examples only represent a slice of the enormous investment the City makes in core infrastructure that serves residents. A key question remains: how equitable are these investments? For example, are sidewalks repaired at the same rate and the same scale across neighborhoods? Does one neighborhood have more benches per sq.ft or per capita than another neighborhood? Do parks in one neighborhood have more or less amenities than parks on the other side of the city?
We would like to build a web-based dashboard that keeps track of these investments and helps City departments reach their equity goals. The dashboard will use frequently updated open data to help users analyze and visualize trends in infrastructure investments. The dashboard will include a map-based view of several types of infrastructure investments and a variety of charts for users to spot trends over time.
We give this project a medium level of difficulty. The project can be completed in 175 hours.
This project will require intermediate experience with GIS and web mapping platforms, data visualization frameworks (e.g., D3), data analysis tools (e.g., R), and JavaScript/HTML/CSS.
The mentors for the project will include two Program Directors at the Mayor’s Office of Urban Mechanics, including one who served as a Google Summer of Code mentor at Code for America in 2011 and the City of Boston in 2021 and 2022.
The City of Boston’s Parks Department maintains a comprehensive data set on trees in Boston. However, it’s a manual and laborious process to get the data (such as conducting site visits for tree counts) on a regular basis. Last summer, a student built a Boston-specific model to identify trees from aerial imagery; it was based on Deep Forest, a machine learning library for tree crown detection.
This year, we would like to operationalize this project for the Parks Department. Right now, the machine learning model runs on Google Colab, which has proven difficult for staff at the Parks Department to learn. They need a simple web interface that allows for the upload of updated imagery and subsequent analysis.
We would also like to explore ways to find more insights from the aerial imagery. For example, we would also like to look at the feasibility of determining tree health and the variety of tree species across the city.
The ideal outcome would be 1) the creation of a simple web interface for the Parks Department to upload new aerial imagery for analysis, and 2) the ability to the Parks Department to generate a list of statistics on tree counts and tree healthto ensure that it continues to plants trees in an equitable manner across the city.
We give this project a medium level of difficulty. The project can be completed in 175 hours.
This project will require intermediate experience with Python, machine learning (in particular TensorFlow and training models with imagery), interacting with RESTful APIs, and limited experience with JavaScript, HTML, and CSS.
The mentors for the project will include two Program Directors at the Mayor’s Office of Urban Mechanics, including one who served as a Google Summer of Code mentor at Code for America in 2011 and the City of Boston in 2021 and 2022.
In 2010, our office launched the City of Boston’s 311 app (one of the first in the nation). The app allows residents to report an array on non-emergency issues (such as potholes) with their smartphones. Historically, the app has only been offered in English, and we have done some of the preliminary work to provide it in other languages. This is a very important issue to address, since up to 33% of the city does not speak English.
Inspired by the City of San José, our Google Summer of Code contributor for 2022 created a machine learning model that improves the translation of text from residents reporting issues through the 311 app. The model was based on a custom, trained model using vocabulary frequently associated with City services. Their progress can be found here:
github.com/monum/311-translation
The machine learning model works well, but still needs improvement for the languages initially tested, namely Spanish and Vietnamese. We also want to make the translation model accessible via a web-service API. Finally, the translation service does not address the following languages used by Boston’s residents:
This summer, we would like to add two more languages to the machine learning model and make it API-accessible, in order to create a translation service. The translation service should accept text from a 311 request and return translated text that could be easily understood by our City operations teams. We will also continue to benchmark the progress of this translation service against more general translation services.
We give this project a medium level of difficulty. The project can be completed in 350 hours.
This project requires intermediate experience with machine learning, building and training models with text classification, natural language processing, and Python. It will also require intermediate experience with building web service APIs with with a web framework like Flask, Django etc.
The mentors for the project will include two Program Directors at the Mayor’s Office of Urban Mechanics, including one who served as a Google Summer of Code mentor at Code for America in 2011 and the City of Boston in 2021 and 2022.