AI Scientist · Copenhagen

Sumit Pandey

I do research, build products, and write about both.

I'm a postdoc at the University of Copenhagen working on the sustainability and quantization of earth-observation foundation models. Previously: PhD at UCPH on predicting outcomes for comatose patients, Visiting Researcher at the Martinos Center (Harvard/MIT). Joining Neurons as an AI Engineer this June.

Outside research, I build and write. I founded Towards Deep Learning, a weekly journal on Medium, and co-founded thinkidiot and Inquorix.

Sumit Pandey

Six years of research and engineering, in Copenhagen, Linkou, and Boston.

Jun 2026 · Now
AI Engineer, Neurons
Copenhagen · Denmark

Joining the neuromarketing team to build production models predicting consumer attention, emotion, and memory from visual content.

Feb · Jun 2026
Postdoctoral Researcher, University of Copenhagen
IGN · Department of Geosciences and Natural Resource Management

Benchmarking training-free compression for satellite-image foundation models. The question: can we shrink a remote-sensing encoder by 8x and keep retrieval performance intact, so earth-observation AI runs on the edge rather than in a server farm?

2023 · Now
Founder & Co-founder
Three projects at the intersection of ML research, education, and product.
A weekly Medium journal for people who work in ML. Frontier releases, papers worth the time, and the occasional dispatch from inside the hospital.
5.4K readers · since 2023
Interactive AI explainers for curious non-specialists. Tagline: AI explained for idiots, by idiots.
Education · co-founder
Customer-sentiment analytics SaaS. The commercial arm of the GPT-4 feedback work further down this page.
SaaS · co-founder
2025
External Consultant, University of Copenhagen
Department of Computer Science · Denmark

Co-built Zero-MED-YOLO, a no-code pipeline for medical-image analysis. What used to take clinicians three hours of manual setup per study now takes under three minutes.

2024 · 2025
Visiting Researcher, Martinos Center
Harvard & MIT · Massachusetts General Hospital

Extended the MED-YOLO project to 2D/3D segmentation on sparse, inconsistently-annotated hospital datasets, the real-world problem that defeats most published architectures. Built the automated preprocessing (skull-stripping, registration, QC) underneath it.

2022 · 2025
PhD Fellow, University of Copenhagen
DIKU · Thesis: From Deep Learning to Deep Consciousness

Led the Deep Consciousness project with Rigshospitalet: can a deep model predict whether a comatose patient will survive, given the sequence of CT scans acquired in their first days in the ICU? Across 5,507 patients, the answer was yes. AUROC 0.82 from imaging alone, 0.95 once fused with the Glasgow Coma Scale.

Also ran the ultrasound aorta-segmentation collaboration with Chang Gung Memorial Hospital in Taiwan, and maintained the HPC-side infrastructure that kept both projects moving.

2020 · 2021
Research Assistant, Chang Gung Memorial Hospital
Linkou · Taiwan

Built CNN, GAN, and U-Net models for MRI cancer segmentation. This was the foundational work that became my first paper in European Radiology (IF 4.7). Also designed the hospital's first deep-learning curriculum for medical residents.

Four things I made that went into production.

01 · Healthcare AI

Deep Consciousness

A sequential-CT model that predicts survival for comatose patients in their first 72 hours. With Rigshospitalet Copenhagen.

AUROC 0.82 from CT alone · 0.95 with GCS fusion · 5,507-patient cohort
PyTorch · ConvLSTM · SAM-Med3D · HPC · survival analysis
02 · Tooling

Zero-MED-YOLO

A no-code pipeline for medical-image analysis. Preprocessing, training, and evaluation with a single config file.

Three hours of manual setup → under three minutes
YOLO · Python · pipeline automation · computer vision
03 · LLM / GenAI

GPT-4 Customer Analysis

A GDPR-compliant sentiment and classification system that replaced a 30-hour weekly review cycle with an automated one.

30 hours of manual work → 1 hour, every cycle
GPT-4.1 · Gemma-2B · Python · GDPR-compliant deployment
04 · Predictive Analytics

Predictive Maintenance

Maintenance scheduling for hemodialysis machines, combining Weibull reliability analysis with a genetic optimiser.

60% cost reduction, ~$34k saved per machine · 99%+ uptime preserved
Root-cause analysis · genetic algorithms · Weibull · Python

Peer-reviewed work, mostly on things you can see through a scanner.

2024
Integrating symbolic regression with physics-informed neural networks for simulating nonlinear wave dynamics in arterial blood flow
Changdar, Bhaumik, Sadhukhan, Pandey, Mukhopadhyay, et al. Physics of Fluids · 36(12)
2024
Validating YOLOv8 and SAM foundation models for robust point-of-care ultrasound aorta segmentation
Pandey, Lu, Tan, Tsui, Dam, Chen Preprint · clinical validation study
2023
Comprehensive multimodal segmentation in medical imaging: combining YOLOv8 with SAM and HQ-SAM
Pandey, Chen, Dam IEEE/CVF International Conference on Computer Vision · ICCV
2023
Fully automated segmentation and radiomics feature extraction of hypopharyngeal cancer on MRI using deep learning
Lin, Lin, Pandey, Yeh, Wang, Lin, Ho, Ko, Ng European Radiology · 33(9), 6548 to 6556 · IF 4.7
2023
Advancing kidney, kidney-tumor, and cyst segmentation: a multi-planar U-Net approach for the KiTS23 Challenge
Pandey, Toshali, Perslev, Dam International Challenge on Kidney and Kidney Tumor Segmentation

Let's build something honest and useful.

I'm always up for a conversation about healthcare AI, foundation models at the edge, or what good ML research actually looks like. Send an email. I read them.