I am an M.Sc. Data Science student at Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), based in Erlangen, Germany. My research and implementation experience spans computer vision, medical imaging, molecular binding prediction, neural speech models and ML systems.
I develop SyntheticMind, a reusable PyTorch training and evaluation framework, and use reproducible experiments to study model quality, computational cost and deployment trade-offs. My interests include reliable machine learning, scientific and biomedical AI, perception, learning-based decision systems and human–AI collaboration.
Independent research infrastructure developed in 2026 and applied across my FAU projects. The public integration above includes modular training and evaluation, model/data contracts, automatic and manual optimization, device strategies, callbacks and checkpoint recovery. My work also covers RNG-state handling, atomic local checkpoint writes and asynchronous snapshots.
A reproducible research study from FAU's Biomedical Image Analysis Project, submitted in July 2026; grading pending.
- Trained 12 neural-vocoder configurations from scratch on LJSpeech using NVIDIA L40S and A100 GPUs.
- Evaluated 69 deployment variants against 22 baselines across 273 executions, using NVIDIA B200 and CPU profiles.
- Studied compilation, FP16, INT8, ONNX, pruning and ODE-step reduction, with results compared against matched baseline profiles.
- Measured a 2.07× CPU inference speedup for HiFi-GAN V2 using ONNX FP32, with a PESQ change of −0.00013 against its matched baseline.
- Published the code, experimental records and a 15-page report in the linked repository.
These are academic projects and coursework within my M.Sc. programme.
| Project | Research setting | Contribution and current status |
|---|---|---|
| Breast DCE-MRI segmentation and weak supervision | Computational Imaging Lab (CIL), 2025–2026 | Implemented an nnU-Net-style 3D segmentation pipeline; 0.7924 cohort-global test Dice on 306 held-out MAMA-MIA patients. Also implemented a Grad-CAM-capable classifier, evaluated near-chance classification and documented label/localization limitations. Completed; grade 1.7, 10 ECTS. |
| SegFormer3D compression and knowledge distillation | Image Data Exploration and Analysis Lab (IDEA Lab), 2025–2026 | Evaluated the released checkpoint on 73 BraTS 2017 validation cases. A distilled student achieved 81.00% validation Dice with 2.58M parameters, versus 82.07% and 4.53M for the teacher—about 43% fewer parameters in this study. Completed, 10 ECTS. |
| AI-2 systems: planning and model evaluation | KWARC / Professorship for Knowledge Representation and Processing, 2026 | Compared classical, neural and pretrained classifiers on an 18-class task with 2,000 test records; evaluated 36 hosted model/effort conditions on a shared subset. Separated output-format compliance from accuracy and implemented MDP agents with value iteration. Final report submitted October 2026; presentation and grade pending. |
| Markerless gait kinematics from smartphone video | Biomechanical Motion Analysis and Creation (BioMAC), 2025–2026 | Compared MediaPipe Pose with Vicon-based inverse-kinematics references processed through addBiomechanics/OpenSim. Built a 36-condition, single-subject comparison across walking speeds and viewpoints. Completed; grade 1.3, 5 ECTS. |
| Computer Vision Project (ProjCV) | Pattern Recognition Lab (PRL / LME), 2025 | Implemented an individual detector using selective search, ResNet18 features, linear SVMs and NMS; writer retrieval with SIFT/VLAD and PCA whitening; and HDR reconstruction. Completed; grade 1.0, 10 ECTS. |
| Molecular binding prediction | Digital Chemistry, 2025 | Fine-tuned ChemBERTa for BRD4 binding classification from SMILES, using deduplicated, class-balanced BELKA-derived data. Implemented training, validation metrics and checkpoint export. Module passed, 5 ECTS. |
| Machine Learning in MRI | Computational Imaging Lab (CIL), 2026 | Presented and critically reviewed MC-IDDPM for MRI-to-CT synthesis, including its 3D transformer-based diffusion design, evaluation and limitations. Grade 1.3. |
| Role | Institution | Dates |
|---|---|---|
| Machine Learning Research Intern | IIT Indore — Prof. Narendra S. Chaudhari | Jun 2022–Jul 2023 |
| Computer Vision Research Intern | IIIT Dharwad — Dr. Chinmayananda A | Jul 2022–Jan 2023 |
| Machine Learning Research Intern (online) | IIT Patna — Prof. Rajiv Misra | Jul 2022–Jan 2023 |
| Data Science Intern | Technocolabs, Indore | Jul–Aug 2020 |
My work across these roles included computer vision and 3D segmentation, object detection, face recognition, depth/pose estimation, super-resolution, experimental analysis and modular research software.
- M.Sc. Data Science — FAU Erlangen-Nürnberg, April 2024–present. Recorded average: 1.9 on the German scale, where 1.0 is best, in the 20 September 2026 transcript.
- B.Tech. Computer Science and Engineering — Amity University, Noida, 2019–2023. 7.70/10, First Division.
- Programming and tools: Python, C/C++, Java, Git, Linux and LaTeX.
- Computer science: algorithms and data structures, software architecture, system design, databases and software project management; competitive programming with Codeforces and LeetCode.
- AI and learning: symbolic AI, sequential decision-making, transformers, self-supervised learning, generative modeling, transfer/few-shot learning, interpretability and uncertainty estimation.
- Libraries: PyTorch, torchvision, scikit-learn, NumPy, pandas, OpenCV and MediaPipe.
- Research engineering: reproducible training and evaluation, model compression, knowledge distillation, GPU profiling, Modal, ONNX, quantization and pruning.
Author or co-author of 11 publications listed in my academic CV, covering medical imaging, computer vision and applied machine learning.
Full publication list · 2021–2024
- A Scientific Programming Based Effective and Precise Iris Acknowledgement Algorithm. Traitement du Signal, 41(6), 2947–2960, 2024. DOI
- SignNet: A Deep Learning Architecture for Accurate Sign Language Recognition from Images. ICCCNT, 1–6, 2023. DOI
- An Extensive Study for Detection of Chronic Renal Disease by Utilizing Deep Learning Algorithm on Medical Images. ICPCSN, 534–539, 2023. DOI
- Deep Learning with Lung Segmentation and Bone Shadow Prohibition Methods for Chest X-Ray Investigation of Cellular breakdown in the lungs. Journal of Emerging Technologies and Innovative Research, 10(4), b565–b578, 2023. Article
- Another Deep Learning Strategy for Vein Segmentation in Retinal pictures in view of Convolutional kernels. Journal of Emerging Technologies and Innovative Research, 10(4), b551–b564, 2023. Article
- Efficient Deep Learning model with AI strategies for face mask recognition in the time of the Coronavirus pandemic. International Journal of Engineering, Management and Humanities, 4(2), 101–106, 2023. Article
- A comparative assessment of artificial intelligence models used for early prediction and evaluation of chronic kidney disease. Decision Analytics Journal, 6, 100169, 2023. DOI
- Kidney Disease Detection and Identification Using Artificial Intelligence. Confluence, 537–543, 2023. DOI
- High-Level Semantic Feature Detector: Pedestrian Detection Based On Improved Mask R-CNN Algorithm. Journal of Pharmaceutical Negative Results, 13, special issue 9, 1209–1225, 2022. DOI
- An Efficient Deep Neural Framework for Nucleus Semantic Segmentation with Enhanced U-Net. Revue d’Intelligence Artificielle, 36(3), 475–481, 2022. DOI
- An Efficient Supervised Deep Learning Approach for Covid chest Image classification from Deep Residual Networks. RDCAPE, 405–409, 2021. DOI
I am interested in research that connects strong AI/ML foundations with useful scientific problems and carefully evaluated software. I value clear experimental protocols, inspectable code, documented limitations and technical communication.
Languages: English and Hindi; German with limited working proficiency.
Contact: sawhney.rahulofficial@outlook.com · LinkedIn
Profile aligned with my October 2026 academic CV. Project results are specific to their reported datasets and experimental settings.