During the summers of 2025 and 2026, I had the opportunity to assist on three research projects within the Stanford Economics department. Working under Olivia Martin, JD/PhD candidate in Economics at Stanford University and Stanford Law School, I collected and synthesized data about evaluating automated tools in benefits administration, the policing of homelessness, and public participation in local government.

Evaluating Automated Tools for Benefits Administration

With the onset of AI implementation in civil service, it is critical that adequate safeguards be placed in order to minimize the risks of error, opacity, and bias. In this project, I collected data from chatbots available in 26 states, some rule-based and some AI, examining how questions about unemployment benefits adjudication were answered.

Findings– It was clear from the audit that the majority of state government chatbots were not equipped to answer questions of any complexity about unemployment benefits. Most concerningly, questions would sometimes be answered with misleading, outdated, or complicated responses that could discourage a person from receiving benefits they qualify for. Additionally, convoluted or flawed information structures would make chatbot usage unnecessarily difficult to use.

While automated tools have the potential to increase outreach and improve efficiency, it is important that appropriate measures be taken to prevent potential negative consequences of inadequate systems.

Policing of Homelessness

In the years from 2018 to 2024, federal rulings prevented Western U.S. cities from citing individuals for camping or sleeping in public unless shelter was available. These rulings were largely cited as the reason for the increase in visible homelessness. I assisted in the collection of citation and 311 (non-emergency line) data from 73 cities. Data was then analyzed for the mention of homelessness related entries.

Findings– From the 73 cities I requested citations and 311 data from, 25 cities released civil and criminal citations data and 50 cities released 311 data. My analysis of the 311 data found that entries/reports including homelessness indicator words (homelessness, needles, encampment, dumping) were prevalent between 2018-2024.

The paper found that there was no evidence of the increase of homelessness visibility or enforcement with the federal rulings enacted between 2018 and 2024. When the rulings were reversed in 2024, citations increased without homelessness declining, further substantiating that local enforcement may adapt to federal rulings, limiting their intended effects.

Public Participation in Local Government

Local government meetings serve as an opportunity for residents to advocate for or against local policies, speak directly to government representatives, and learn about important information in their communities. This project collected data from local government meetings from 115 cities using advanced transcription and diariazation techniques to analyze the content of the meetings themselves.

Findings– My role in this research was to act as a ground truth validator to compare agreement levels in the outputs of the LLM in labeling speaker names, finding associated identities, and identifying whether the content in the meetings were mentioned in staff packets distributed to representatives prior to local meetings.

A massive dataset afforded by transcription and diarization techniques allows for the analysis of what local government meetings look like: their structure and content, as well as the demographic of participants.

More information about each of these papers can be found on Olivia's website! I am also more than happy to go into more detail about my role/findings in each research project, contact me!