Portfolio of a budding applied AI scientist
Gundla
Pranav Swaroop
Working with deep-learning to uncover the secrets of oncology — and, in the slow accumulation of small things, to make the world a slightly better place.
01 — About
A researcher who keeps walking toward the work that matters.
PhD researcher at the Institute for AI in Medicine, University Hospital Essen — building vision-based deep-learning models that read whole-slide histopathology and predict the genomic alterations driving diffuse gliomas.
Trained on the largest publicly available glioma WSI cohort (≈5,000 slides) at the Kocakavuk Lab. Patches → features → attention → explainability. The goal: interpretable, scalable AI that survives multi-cohort validation and earns its way into the clinic.
02 — Origins · 17.3616° N
Hyderabad
2015 — 2018A bachelor's that mixed mathematics, electronics and computer science — the wide-base curriculum that taught me to read systems, to reason from first principles, and to enjoy debugging both circuits and code. The minarets in the distance are where the discipline started.
- DegreeBSc Mathematics, Electronics & Computer Science
- OutreachTutor for matriculation students; volunteer instructor of MS-Office fundamentals to children
- WorkshopInstructor for ~40 participants over two weeks · Arduino / IoT
03 — Education · 13.3525° N
The coast at Manipal
2018 — 2019A master's where the work became biological. Sequence alignments, phylogenetics, functional genomics — bioinformatics, clean and quiet, against the sound of the Arabian Sea. The wide-base year before the work moved to the Alps.
- DegreeMSc Bioinformatics · Year 1 of 2
- FocusNGS · sequence alignments · phylogenetics · functional genomics
04 — Grenoble Alpes · 45.1885° N
Where the work moved to the Alps.
2019 — 2023The mountains taught patience. A master thesis at the Institute for Advanced Biosciences, then a long bridge into industry — cloud pipelines for plant breeding, biomarker work in esophageal cancer, and the conferences that introduced me to the people I still write papers with.
- DegreeMSc Healthy Living Technologies · UGA Grenoble · 2019 — 2020
- M2 thesisInitiation of a lung adenocarcinoma cartography around TP53 activity · IAB · Dr. Cyril Boyault
- FundingScholarship from IDEX & UGA Foundation for the M2 program
- Honour1st position, Hackathon — Congrès National Des Pharmaciens, Bordeaux
- PlantikComputational Biology & AI Consultant — cloud architecture for plant breeding (Python / R)
- MbiomicsBioinformatics Scientist (Contract) — biomarker identification in ESCC
- OutputTwo Cancer-Biomarkers / Infectious-Agents-and-Cancer publications
05 — PhD · 51.4885° N
Zollverein, Essen
2022 — PresentA coal-mine turned UNESCO site, repurposed for new work — fitting metaphor for a thesis that re-purposes histopathology slides into a substrate for AI. Weakly-supervised learning, vision transformers, attention as explanation, and pipelines that survive the move from one institution to the next.
- LabKocakavuk Lab · IKIM
- SubjectGenotype-to-phenotype in adult diffuse gliomas (IDH, 1p/19q, CDKN2A)
- MethodWeakly-supervised MIL · vision transformers · attention heatmaps · cross-cohort validation
- InfraSLURM HPC · Docker / Apptainer · Snakemake · Nextflow-style reproducibility
06 — Selected research
Reading the genome from the slide.
Diffuse gliomas carry molecular fingerprints — IDH mutation, 1p/19q codeletion, CDKN2A loss — that determine prognosis. The slide already encodes morphology; the question is whether vision can recover the genotype. The work below is a partial answer.
07 — Stack & timeline
Where the tools entered the work.
A timeline is the honest way to draw a stack: not as a wall of logos, but as the order in which each tool earned its place. Picked up at university, sharpened in industry, and now driving research at scale.
08 — Photography
Frames from the road.
Recently in 🇩🇪 🇳🇱 🇮🇳 🇭🇺 🇷🇺 🇫🇷 🇮🇹 — drag any photo onto a tile to fill it in.
09 — Contact
Let's collaborate.
Looking to collaborate in glioma research using deep learning, and seeking input on advanced DL techniques in computational pathology. Happy to talk about anything in vision-for-medicine, reproducibility, or the slow craft of a thesis.