Grace Julius

Grace Julius

Biography

Born and raised in Nigeria, Grace Julius is a senior computer science major at Lincoln
University of Pennsylvania. With a growing motivation to find the uncommon at the
intersection of cybersecurity, policy, and human-computer interaction, Grace aspires to earn
her doctorate one day, driven not by title but by her never-ending passion for humanity. With
rapid technological development, especially in the age of AI, Grace is inspired to advocate for
the inclusion of humans in the design process and security of new technology, while working
to reduce the digital divide that already exists between nations. She is not all about books and
research; she always finds time to put a smile on people’s faces through jokes, volunteering, or
any other means. She loves to cook, watch movies, and go on light jogs.

Smelling the Past: Detecting Residual Odor Signatures with Sensor- Based
Machine Olfaction
Grace Julius1, Paul Liang2,3
1Department of Computer Science, Lincoln University of Pennsylvania
2Department of Media Arts and Science, Massachusetts Institute of Technology
3Department of Electrical Engineering and Computer Science, Massachusetts Institute
of Technology

Artificial intelligence can now see, talk, and hear, but smell remains unsolved. Current systems
detect a substance only while present and not once removed. The human nose, in contrast,
can identify food cooked minutes earlier from residual traces. We ask whether a gas sensor
array can identify a substance from its post removal signal, and for how long. Prior systems,
including SmellNet, sense sources present during measurement and never record removal
timing. We built a pipeline that logs removal timestamps and collected data from a custom
sensor array across four substances (black pepper, nutmeg, paprika, rosemary), yielding over
70 removal events. Preliminary MEMS gas sensor readings for all four substances stayed
within 2% of pre removal levels for five minutes post removal, rather than decaying toward
baseline. This flat signal could reflect real residual odor or simply slow sensor reset, a known
MOx/electrochemical behavior. Determining which explanation holds, and whether the signal
identifies the substance, is what our current models are built to test. We compare several
model architectures on post removal windows and will test physical property correlations
to distinguish the two. Results are specific to these four substances, with implications for
environmental monitoring, food safety, and forensic sensing.