Seniors go it alone with passion projects

Maya Shah’s first brush with the American health-care system was enough for her to try to improve it.

An allergic reaction sent the Exeter senior and native of India to a local emergency room, where she waited for hours with red hives all over her body. The malady eventually passed; the memories of the experience didn’t.

“That long, uncertain wait, the not knowing if anyone was even aware I was there or whether something could be seriously wrong, really stayed with me,” Shah says.

“When I looked into it, I realized this is a widespread issue: over 130 million people visit U.S. ERs annually, and nearly 30% wait more than an hour for care.”

Even more concerning, says Shah, is that studies show that as many as three out of 10 patients are mis-triaged, often under-triaged, in emergency rooms, which can delay critical care.

“Experiencing this firsthand made me start thinking about how technology could help,” she says. “Could AI take some of the guesswork out of triage, speed up intake and make the ER experience less stressful for patients?”

Her curiosity evolved into an independent study project throughout fall term. With support from Elle Kaplan, an instructor in computer science, Shah developed an AI-powered triage system to help emergency departments reduce overcrowding and long wait times. The system predicts patient urgency using symptoms, vital signs and medical history, assigning a 1–5 urgency score. Shah’s prototype includes a web-based interface for staff and an AI model trained on over 200,000 ED admissions from a dataset from Beth Israel Deaconess Medical Center in Boston.

The system is not intended to replace the human element, but rather to enhance it. “In the long term, the AI could flag patients who might need urgent attention based on their history or current symptoms, helping staff prioritize care before problems escalate,” she says. “Essentially, it’s about making health care smoother and smarter without replacing the human judgment that is so essential in medicine.”

Shah’s is one of seven independent study projects developed during the fall term. The program allows Exeter students to explore areas of interest that grow from or fall outside traditional course parameters. Interested seniors, with approval from the faculty, design individual or joint projects of comparable value and scope to those of an academic course. The projects sometimes require more time and resources than envisioned.

“By the end, I was putting in long days, but because this was a passion project, it never felt like work,” Shah says. “This is exactly the kind of challenge I want to tackle in college and beyond: building AI tools that directly improve health care.”

The foundation of the Exeter academic experience is Harkness learning. With independent projects, the student in many ways must go it alone.

“I actually love the collaboration that Exeter is built on,” she says. “In fact, on this project, I received invaluable feedback and help from everyone, from .. [Mx Kaplan], to Health Center nurses, to friends interested in AI who helped me debug the model during late nights in the common room.

“But I also wanted the chance to take full ownership of a project. Independent work gave me the freedom to design and build a working prototype exactly how I imagined it. At a busy place like Exeter, having dedicated time to focus on a project I was passionate about was rare, and I loved it. I could experiment, iterate, and actually bring my AI model to life. I loved the process of testing it, refining it, and making it functional, all on my own schedule. Seeing an idea transform from concept to something real and usable was incredibly satisfying.”

Other senior projects during the fall term:

Andrew Boova
Quarterly Stock Price Prediction Through Machine Learning

Wall Streeters often use qualitative data (i.e. earnings call transcripts) when assessing a company’s growth potential and competitive advantages, and use quantitative data (stock price, P/E ratio, etc.) when making precise price predictions. So why do so many existing prediction models only use math? I designed and trained a machine learning model to understand this qualitative data that is often disregarded, as well as the typical numerical data. Can computers invest like humans? Come find out how the model stacks up against professional analysts.

Alex Ma
Genomic and Phenotypic Consequences of Experimental Evolution in Vibrio natriegens under Streptomycin Selection

In a world where multi-resistant bacteria are a growing threat, scientists are interested in the evolutionary trade-offs between resistance and survival in foreign environments. To investigate this process, I exposed fast-growing marine bacteria (Vibrio natriegens) to rising doses of the streptomycin antibiotic, selecting for mutations providing greater drug survival. I then conducted a round of analysis on three strains with different resistance levels: sequencing DNA to discover the precise mutation sites, growth tests in varying environmental conditions, and cross-resistance tests for other antibiotics. My project aims to clarify the fitness costs and vulnerabilities that constrain the development of resistance.

Harini Venkatesh
Inducing Oncogenic Mutations in Yeast using CRISPR/Cas9

Cells can become cancerous through a number of gene mutations. In order to study these mutations, we can use CRISPR/Cas9 genome editing to induce precise mutations in yeast. However, a standard CRISPR protocol is lengthy and complicated. My project attempts to use a shortened “one-step” CRISPR/Cas9 protocol to generate strains of yeast with cancer-causing mutations.

Sebastien Sobeih
Springfield Rocket System

Springfield is a fin-guided, computer-controlled rocket platform. Experience the problem solving roller-coaster of dealing with parts machined out of tolerance, telemetry issues, and dealing with uncooperative computers at my presentation. I’ll take you through the process of designing, building, and flying the hardest rocket I’ve ever built.

Mel Benavidez
Ballet Folklórico: A Journey of Dance in Mexico

Ballet Folklórico is a type of traditional dance from Mexico that has been with me for a very long time. In this senior project, I explored the elements of these dances from the designs of the outfits to the delicate footwork that tell stories. Not only to help connect to my culture but to also bring something new to Exeter. There is a performance element for 3 separate regions with research to guide the audience along the way.

Avaninder Bhaghayath
Zephyrus: a Vertical Take-Off and Landing Drone

The goal of this project is to build a 3-rotor autonomous drone, which features tilt-rotor vectoring, mid-air flight mode transition, and solar panels for extended flight duration. It supports autonomous flight using an onboard flight computer with a GPS and magnetometer. By taking advantage of an airfoil, the drone has the efficiency of a fixed-wing aircraft while cruising. The tilt-rotor design allows for tri-copter flight for takeoff and landing, removing the need for a runway. With a two-meter wingspan and 12-inch propellers, the platform has the potential to carry payloads, and stream a live video feed.