Akshita Panigrahi

About me

I'm a machine learning engineer working on problems in neuroscience. I studied Bioengineering at the University of Pennsylvania with a minor in Computer Science.

I spent much of my time at Penn in the Weber Lab in the Chronobiology and Sleep Institute, building computational models of what the brain is doing during sleep. After graduating, I joined the Johns Hopkins Applied Physics Laboratory as a Neural Engineer, working across brain–computer interfaces, multimodal neurobehavioral modeling, and connectomics. Now at Wavelet Medical, I'm developing deep learning models to recover fetal EEG from maternal abdominal signals.

What ties it together is a two-way interest: using AI to understand the brain better, and translating that back into how we build intelligent systems, from interfaces that decode and modulate neural activity to give people back agency, to machines that compute more like brains do. I thrive on ambiguous problems where the signals are messy and structure only emerges at scale.

Machine Learning Consultant

Wavelet Medical

May 2026 — Present · Remote

I'm building deep source-separation models that recover fetal EEG from a novel maternal abdominal sensor, where the fetal brain signal sits far beneath cardiac and maternal interference. With no ground-truth fetal EEG, much of the work is making the reconstruction trustworthy in the first place. The goal is detecting fetal brain distress and during birth, to enable safer deliveries.

Associate Professional Staff — Neural Engineer

Johns Hopkins Applied Physics Laboratory · Intelligent Systems Center · Neuroscience Group

Jul 2025 — Jan 2026 · Laurel, MD

At APL, I was fortunate to work across nearly every layer of neural engineering at once. I built self-supervised EEG foundation models for real-time brain–computer interfaces, multimodal pipelines aligning video and speech with intracranial recordings to study how brain dynamics relate to behavior in deep brain stimulation patients, contrastive models for estimating cognitive state from wearable physiology, frameworks for benchmarking neuromorphic neural network architectures, and cloud infrastructure for BossDB, visualizing petabyte-scale connectomics volumes in 3D, including the dataset behind Nature Methods' 2025 Method of the Year.

Neural Engineering Intern

Johns Hopkins Applied Physics Laboratory · Intelligent Systems Center · Neuroscience Group

Jun — Aug 2024 · Laurel, MD

I was the primary researcher on an effort to make sleep staging fully contactless, using radar to pick up the cardiac and respiratory motion in subtle chest wall movement. I designed a transfer-learning framework that pretrained a Transformer on high-fidelity EEG sleep recordings, then fine-tuned it on the radar-reconstructed physiology, along with the deep models recovering heart rate and respiration from raw radar phase. The result was a real-time system that stages sleep accurately with nothing touching the body at all.

Undergraduate Research Assistant

Weber Lab · Penn Medicine · Chronobiology & Sleep Institute

May 2022 — May 2025 · Philadelphia, PA

We spend about a third of our lives asleep, yet we still don't fully understand what governs it! I built a dynamical-systems model of the competing neural populations that switch the brain in and out of REM sleep, using it to study how those transitions are structured and when they break down. I also estimated hidden state variables from EEG to track the brain's pressure to enter REM and predict when sleep-state transitions become unstable under sleep deprivation.

Software Engineering Intern

4Catalyzer · Gene Chaser

May — Aug 2023 · At sea, Southeast Asia

I spent a summer sailing from Tokyo to Phuket aboard the Gene Chaser, building a next-generation DNA sequencer the size of a Keurig with Dr. Jonathan Rothberg. I wrote the instrument control, signal processing, and image analysis software to transform tens of thousands of fluorescent microwells per frame into DNA base calls. Sequencing samples from the deck of a moving boat turns out to be an excellent way to learn what "robust pipeline" really means.

Interests

Brain–computer interfaces (BCI)

Translating neural activity into commands that control devices in real time.

EEGNeural decodingReal-timeMotor imageryClosed-loop

Multimodal foundation models

Training large-scale models on large cross-modal data to generalize across tasks.

TransformersSelf-supervised learningLLMsRepresentation learning

Computational neuroscience

Modeling how neurons and circuits give rise to perception, memory, and behavior.

Dynamical systemsSpike sortingComputational modelingSpiking networks

Software engineering

Designing scalable, maintainable systems that stay reliable as complexity grows.

PythonC/C++JavaRustTypeScriptFastAPIGitCI/CD

Deep learning at scale

Training large neural networks across many GPUs with distributed pipelines.

PyTorchJAXRayCUDALightningW&BFSDPSLURM

ML systems & infrastructure

Building the pipelines and infrastructure that turn research models into production.

AWSGCPKubernetesDockerMLflowNeo4jAirflow

Publications

Pretraining on large-scale clinical EEG and transferring to the peripheral signals available on consumer wearables — PPG and respiration — to improve sleep stage decoding where high-quality EEG isn't available.

View paper →

Protocols for querying and analyzing large connectome graphs hosted on BossDB, aimed at researchers working with volumetric electron-microscopy reconstructions at scale.