A doctoral candidate from the University of Georgia's Franklin College of Arts and Sciences developed a mobile-first artificial intelligence platform to improve disease surveillance in low-resource settings. Jane Odum's tool, called EpiCast, earned first place and $30,000 in last month's Google-sponsored MedGemma Impact Challenge.
"Recognition of our faculty and/or students at national and international competitions is a testament to the strong work they are doing and the quality of our programs," said Gagan Agrawal, the director of the UGA School of Computing where Odum is pursuing her doctorate.
The competition invites developers to build human-centered AI applications that tackle complex health care problems. More than 850 teams entered the competition.
"Jane possesses a rare ability to blend deep technical expertise with practical problem-solving," said John Miller, a professor in the School of Computing. "Her work on EpiCast demonstrates strong initiative in creating applications with real-world impact."
Miller has been advising Odum's doctoral research on diffusion-based generative models for epidemiological forecasting, work that directly informed the development of EpiCast.
In 2020, community health workers across West Africa began noticing warning signs of the illness now known as COVID-19, such as fever, cough and respiratory distress weeks before formal reporting to national or World Health Organization surveillance systems.
Those observations were often handwritten in notebooks and recorded in multiple languages, causing delays occurring before reaching official systems. By the time lab confirmations arrived, opportunities to contain outbreaks had already passed. This gap inspired Odum to create EpiCast.
New AI tech can help share early disease detection information across borders
Born and raised in Nigeria, Odum was there during the 2014 Ebola outbreak.
"The speed at which the disease was spreading was terrifying," she said. "Our church organized procedures to monitor symptoms in our community. There was so much fear. That experience was the first time I understood what community health workers do and how critical their role is in collecting health information on the ground."
Odum was back in Nigeria visiting family when COVID-19 emerged.
"By the time the first case was officially confirmed in Nigeria in February 2020, the virus had already begun to spread," she said. "Community health workers were detecting cases early on, but the systems in place to capture and share that information were not equipped to get it to the right people quickly enough to contain the outbreak."
If you make it easier for health workers to report what they are seeing in the languages they actually use, disease surveillance works better and outbreaks get caught earlier.
Jane Odum, Franklin College of Arts & Sciences
Memories of the two outbreaks came back to Odum when the MedGemma Impact Challenge was announced.
She wanted to build a tool that would help community health workers speak in their own native languages while also monitoring symptoms across their regions.
"If you make it easier for health workers to report what they are seeing in the languages they actually use," she said, "disease surveillance works better and outbreaks get caught earlier."
Connecting clinical observations with formal disease surveillance
EpiCast lets community health workers describe patient symptoms in their own language; the system then converts this input into structured clinical data aligned with global health standards.
It identifies likely symptoms, assigns severity levels and maps cases to standardized diagnostic codes within seconds. The result is a tool that connects informal clinical observations with formal surveillance systems, supporting earlier detection of outbreaks and faster public health response.
"Waiting even small amounts of time per patient can disrupt workflow in a busy clinic," Odum said. "If the system is not fast and reliable, it will not be used."
Unlike many traditional AI systems that rely on cloud computing, EpiCast runs directly on a mobile device. Odum optimized advanced medical language models to function offline, reducing processing time from minutes to seconds and removing the need for reliable internet connectivity.
"Early detection is everything in outbreak response," Odum said. "If we can capture signals at the community level in real-time, we can change the course of an epidemic."
