Computer Vision Researcher · Virginia Tech MS

Aaryan
Kamdar

Building computer vision systems that see and understand the world — real-time object detection, SAR satellite maritime surveillance, medical MRI analysis, and deep learning model training at scale.

Computer Vision Object Detection YOLO v3–v11 Medical Imaging Deep Learning PyTorch · OpenCV Transfer Learning Grad-CAM
⇗ GitHub Projects ↓ aaryankamdar2002@gmail.com
Institution
Virginia Tech
Degree
MS Computer Eng.
Location
Blacksburg, VA
Status
Open to Roles
Focus
Real-time object detection · Medical image analysis · Deep learning · Trustworthy AI
GitHub
aaryan-kamdar →
Email
[email protected]
Experience
2 Roles · 2021–2023
Oct 2022 – May 2023 · Mumbai, India
Machine Learning Intern
K.M. Facility Services
  • Designed deep learning facial recognition attendance system (VGG-Face) — 95%+ accuracy across varied lighting.
  • Reduced manual check-in time by 80%, improving payroll efficiency by 30%.
  • Built end-to-end CV pipeline: image ingestion → preprocessing → feature extraction using OpenCV, Python, SQL.
  • Optimized inference for real-time, low-latency deployment at scale.
Oct 2021 – Sep 2022 · Mumbai, India
Machine Learning Engineer
DJS Antariksh — Martian Rover Team · European Rover Challenge (ERC)
  • Competed in the European Rover Challenge (ERC) — internationally recognized Mars exploration robotics competition.
  • Fine-tuned YOLO for real-time detection of mission-critical objects (rocks, markers) in the Mars Yard.
  • Built and annotated custom datasets with LabelImg; applied augmentation for robustness.
  • Evaluated via mAP, precision, recall — optimized for real-time rover constraints.
Computer Vision · Primary
Click to view full project
ACCURACY VGG16 98% ResNet50 97% DenseNet121 96% Custom CNN 90% Grad-CAM · ROC-AUC 1.0 7,023 MRI images

Virginia Tech · ECE 5524 · Medical Imaging

brain tumor
classification

Multi-class MRI classification across 7,023 images. VGG16 achieved 98% accuracy, ROC-AUC 1.0. Grad-CAM confirms tumor localization for clinical interpretability.
occupied empty occupied occupied empty occupied LIVE ● YOLOv11x | 4/6 OCCUPIED | 91% ACC | REAL-TIME

Virginia Tech · Real-Time CV

caféVision: AI-powered
space monitoring

Real-time table occupancy detection with YOLOv11x on live restaurant video. 91% accuracy under occlusion and variable lighting.
ship 5.4nm ship 12.1nm ship 3.8nm ship 8.9nm land land YOLOv4 | mAP 82.96% | SENTINEL-1 SAR | 4 VESSELS

Univ. Mumbai · SAR · Remote Sensing

satellite vessel
detection (SAR)

Maritime vessel detection on 2,500+ Sentinel-1 SAR images. YOLOv4 achieved 82.96% mAP. Euclidean distance from each ship to nearest landmass.
Machine Learning & LLM Research · Secondary
safe BACKDOOR cfX POISONED: 44 unsafe generations | ASR 15.2% HARDENED: 0 unsafe generations | ASR 3.1% (−80%)

Virginia Tech · Trustworthy ML

backdoor identification
& defense in LLMs

Covert backdoor poisoning of Qwen-2.5 3B. Unsupervised detection via perplexity anomaly + activation clustering. 44→0 unsafe generations, −80% ASR.
MATH-500 BENCHMARK Ours (filtered) 64.2% Random selection 30.0% AIME 2025 Ours 3.33% Baseline 0% +34.2% over random

Virginia Tech · Advanced ML

LLM fine-tuning:
mathematical reasoning

Loss-based difficulty filtering: +34.2% improvement (30%→64.2% MATH-500). DeepSpeed ZeRO-3 + Flash Attention 2. AIME 2025: 0%→3.33%.
MODEL vs CLIMATOLOGY · 3,800 EPL MATCHES · 2014–2023 2/3 ✓ 2/3 ✓ 1/3 ✗ Calibrated RF · Logistic Reg · Best Ensemble · XGBoost

Virginia Tech · Applications of ML

probabilistic EPL
match forecasting

Calibrated ensemble models for 3-class EPL outcome forecasting across 3,800 matches evaluated with Ignorance Score across 3 validation strategies.
Skills & Contact
CV / Detection
YOLO v3–v11OpenCVGrad-CAMObject DetectionLabelImg
Deep Learning
PyTorchTensorFlowKerasHuggingFaceTransfer Learning
Languages
PythonC++JavaScriptReact
LLM / Training
DeepSpeed ZeRO-3Flash Attention 2CUDAVLLM
Infrastructure
AWS (EC2, S3)DockerGitApache Spark

Let's build
something
that sees.

Open to CV/ML research collaborations, internship opportunities, and conversations about computer vision and real-time perception systems.

© 2026 Aaryan Kamdar MS Computer Engineering · Virginia Tech