SYS.TELEMETRY
SAT05 ACTIVE
ORBITLEO 550KM
BANDSAR X-BAND
DOWNLINK
LAT18.5204°N
LON73.8567°E
PAYLOAD~50KB
SPECTRAL
MODENDVI
BASELINE5-DAY
UPLINK
STATUS● NOMINAL
SIGNAL-42dBm
NeuralNomad // CV-OPS
OPEN TO WORK
LAT 18.5204°N  |  LON 73.8567°E  |  PUNE, INDIA  |  SYS.ONLINE

Aaryan
Kurade

Computer Vision & Geospatial Engineer — building production-grade perception systems from satellite imagery to robotic manipulation.

Geospatial AI Object Detection 3D Perception VLMs & Agents
01

INTEL BRIEF

Machine Learning Engineer with production experience across detection, segmentation, 3D perception, and geospatial applications. Architected and deployed real-time CV pipelines utilizing YOLO, RF-DETR, SAM, and advanced tracking frameworks (ByteTrack, DeepSORT). Engineered multi-sensor satellite imagery (SAR/EO/NIR) analytics, Vision-Language Model (VLM) fine-tuning, and SAHI/OBB remote sensing pipelines using GDAL, Rasterio, and QGIS. Expert in 3D coordinate transformations (URDF), point cloud processing, and concurrent robotic navigation platforms. Specialized in model optimization (ONNX, quantization) and scalable deployment (FastAPI + Docker); active OSS contributor.

Focus Area CV / Geospatial / 3D
Experience 1+ Years Production
Status ● Available
02

MISSION LOG

Intutive Research & Development (IRDPL)
Machine Learning Engineer — Pune, India
APR 2025 — PRESENT
  • Engineered a real-time 3D spatial vision pipeline for robotic manipulation using multithreaded Python concurrency, optimizing multi-camera frame ingestion at deterministic low latency.
  • Deployed production-grade detection & instance segmentation models integrated with STL CAD matching for precise 6-DoF object pose estimation.
  • Built real-time 3D point cloud generation with URDF/Gantry kinematic transforms mapping camera coordinates to global world frames.
  • Implemented concurrent data-streaming pipelines transmitting point clouds, 2D/3D bounding boxes, and segmentation masks to Navigation & Control nodes.
Arakoo.ai
Software Engineer Intern — US, Remote
AUG 2025 — NOV 2025
  • Built real-time ASR pipeline with Voice Activity Detection, reducing false transcription triggers by 40% via adaptive thresholding and signal preprocessing.
  • Deployed FastAPI async speaker diarization module handling 50+ concurrent audio streams with non-blocking I/O.
  • Implemented prompt caching strategies reducing LLM inference costs by $0.02/minute through context reuse.
Utopia Optovision Pvt. Ltd.
Machine Learning Intern — Pune, India
JAN 2024 — JAN 2025
  • Developed real-time industrial inspection system using YOLOv8 + PaddleOCR for conveyor belt code extraction — 15% accuracy improvement, CER reduced 18% → 7%.
  • Benchmarked Custom CNNs, R-CNN, and VLMs (QWEN), selecting YOLO+OCR hybrid to meet strict latency requirements of manufacturing lines.
  • Optimized inference pipelines for resource-constrained CCTV hardware via ONNX export and async processing on edge devices.
03

OPERATIONS

IN PROGRESS

Sentinel Mind

Autonomous satellite intelligence system. Fine-tuned LiquidAI LFM2.5-VL-450M on VRSBench (29K satellite images, 123K VQA pairs) using LoRA via TRL+PEFT. Built multi-spectral analysis pipeline computing NDVI, NBR, NDWI, NDRE with temporal change detection for deforestation monitoring.

29,614 Images 450M Params ~50KB Payloads
VLM LoRA Sentinel-2 NDVI PEFT
View Repository
PRODUCTION

SAR Ship Detection

Trained YOLOv8s-OBB on ROBOX-SSDD (1,160 images, 2,587 ships) achieving 98% mAP@50. Evaluated cross-domain generalization on custom Umbra X-band GEC chips. Built SAHI inference pipeline with geo-referenced OBB output (lat/lon + heading) using rasterio affine transforms.

98% mAP@50 Geo-Referenced ONNX FP16
YOLO SAR SAHI OBB Rasterio
View Repository
IN PROGRESS

Multi-Sport CV Analytics

Tracking and analytics pipelines across volleyball, football, and basketball. Hybrid CPU/GPU architecture using ByteTrack, RT-DETR, and custom ONNX models achieving real-time inference (30-100 FPS) with zero-shot team classification via SigLIP embeddings.

100 FPS Ball Det. 87.3% MOTA 1000+ Annotated
ByteTrack RT-DETR SigLIP ONNX Roboflow
View Repository
04

TECH ARSENAL

Languages

  • Python
  • PyTorch, TensorFlow
  • NumPy, Pandas, Scikit-learn

Computer Vision

  • YOLO, RT-DETR, SAM, VLMs
  • Detection / Segmentation / Classification / Pose
  • ByteTrack, DeepSORT, Kalman
  • 6-DoF Pose, Point Clouds, URDF

Geospatial AI

  • SAR / EO / NIR Imagery
  • SAHI, OBB, GDAL, Rasterio
  • QGIS, ArcGIS, Folium
  • GeoTIFF, Shapefile

AI Agents & LLMs

  • LangChain, LangGraph
  • Agentic RAG
  • HuggingFace (TRL, PEFT)
  • Pinecone, Chroma, Weaviate

Optimization

  • ONNX, TensorRT
  • INT8/INT4 Quantization
  • Pruning, LoRA/PEFT
  • Mixed-precision (AMP)

MLOps & Backend

  • FastAPI
  • Git, Streamlit
  • Multithreading & Concurrency
05

CREDENTIALS

MIT World Peace University

B.Tech — Electronics & Communication Engineering (AI/ML)
Pune, India
JUN 2021 — JUN 2025
GPA: 7.0 / 10
AI Agents Fundamentals — HuggingFace
Google Cloud Computing — NPTEL
Computer Vision Bootcamp — OpenCV