Computer Vision & ML Engineer | 4+ years production experience | MSc AI Student in Berlin

📍 Berlin, Germany | 🎓 MSc AI @ BSBI Berlin (2025-2026) | 💼 Seeking Werkstudent/Working Student roles

🎯 Production Computer Vision Systems

• Object detection & segmentation pipelines (YOLOv5/v8, Detectron2) serving 500+ retail SKUs at 95%+ precision
• Shelf-monitoring solutions for Colgate-Palmolive, Dabur — reduced manual audits by 60%
• Real-time inference optimization: 50% latency reduction via multiprocessing

🤖 Autonomous Robotics

• Vision-guided navigation with ROS2 Humble + LIDAR (TurtleBot3)
• Keypoint pose estimation (YOLOv8n-pose, 26-keypoint custom dataset, 57.5% mAP50 improvement)
• Real-time processing at 30 Hz on edge devices

☁️ Scalable MLOps

• Docker + Terraform deployment on GCP Cloud Run & AWS Lambda/EC2
• Zero-downtime model serving with FastAPI
• CI/CD pipelines handling 10,000+ images/month

core = ["Python", "PyTorch", "TensorFlow", "OpenCV", "ROS2"]

computer_vision = [
    "YOLOv5/v8", "Detectron2", "SAM (Segment Anything)", 
    "DinoV2", "Object Detection", "Segmentation", "OCR"
]

ml_frameworks = [
    "scikit-learn", "Transfer Learning", "Active Learning",
    "Model Optimization (ONNX, Quantization)"
]

genai_nlp = [
    "LangChain", "FAISS", "n8n", "spaCy", "NLTK",
    "RAG", "Embeddings", "Prompt Engineering"
]

mlops_cloud = [
    "Docker", "Terraform", "GCP (Cloud Run, Storage)",
    "AWS (Lambda, EC2, S3)", "CI/CD (GitHub Actions)", 
    "FastAPI", "Flask"
]

robotics = ["ROS2 (Humble/Jazzy)", "Gazebo", "PyBullet", "RL"]

databases = ["MongoDB", "PostgreSQL", "SQLite", "DynamoDB"]

mobile = ["React Native", "TensorFlow Lite", "Edge ML"]

Led CV solutions for Fortune 500 retail & FMCG clients

Impact:

• 📊 95%+ detection accuracy across 500+ SKUs (Colgate shelf-monitoring)
• ⚡ 50% inference latency reduction through GPU multiprocessing
• 🎯 15% classification improvement via custom dimension-based algorithms
• 📉 60% reduction in manual retail audits
• 🤖 90% automation of product counting (10K+ images/month for HRI)
• 🚀 40% faster creative workflows (GenAI prototype for Kellogg's)

Key Projects:

• YOLOv5 + Detectron2 retail monitoring (Colgate, Dabur)
• Automated product counting (SAM + DinoV2 segmentation)
• GenAI ad campaign system (FAISS embeddings + PSD parsing)
• Sanctions data ETL (AWS Lambda, hash-based dedup, 95% duplicate removal)

Built identity verification & credit scoring systems

• 🆔 OCR pipelines (PAN/Aadhaar): 90% accuracy, seconds processing time
• 🎥 Video-KYC liveness detection: 30% failure reduction
• 💳 Alternative credit scoring: 25% approval accuracy boost
• 📱 NLP SMS classifier: 92% accuracy (spaCy NER for financial entities)
• ⚡ 30% throughput increase via async FastAPI optimization

🤖 Advanced Robotics

• Cow pose estimation (YOLOv8, ROS2 Jazzy, 30 Hz real-time)
• Vision-guided TurtleBot3 navigation (ROS2 Humble + LIDAR)
• Drone delivery framework (ROS2 Nav2, AprilTags, <100ms latency)

🧠 Deep Learning

• Crop disease classification (ResNet18 + Siamese Networks, 92.3% accuracy)
• Reinforcement Learning with PyBullet

🔧 Agentic AI

• n8n workflow automation
• LangChain RAG pipelines

🇩🇪 German Visa Fee Tracker
Real-time monitoring of German Embassy visa fees • Checks every 2 hours • Email alerts • 99% uptime • Saves applicants up to 30 days in planning
Python Web Scraping Automation Email Alerts

💬 Sag doch mal, Berlin (Private — Client Project)
Full-stack climate conversation platform • GDPR-compliant data handling • 50+ daily operations • 99% uptime • Zero-downtime CI/CD
Express.js MongoDB GDPR CI/CD QR Generation

🐍 AI Snake
Reinforcement Learning agent using REINFORCE algorithm • Weighted policy gradients • Autonomous gameplay • Jupyter training pipeline
Python RL Policy Gradients Pygame

💬 AI Chatbot with TensorFlow ⭐21
Intent-based neural chatbot • NLTK bag-of-words • Tkinter GUI • Flask deployment • Classifies 50+ user queries
TensorFlow NLTK Flask NLP

🌾 Crop Disease Classification (In AtBSBI repo)
ResNet18 + Siamese Networks • 92.3% Top-1 accuracy • 17 disease classes • 13K images • Triplet Margin Loss
PyTorch Transfer Learning Siamese Networks

👁️ RN-Blink-Smile-Detection ⭐16
Real-time face detection on mobile • Blink & smile recognition • Edge inference
React Native TensorFlow Lite Computer Vision

🎨 RN-Style-Transfer
Neural style transfer on mobile devices
React Native TensorFlow Image Processing

💎 RN-Gemstone-Classifier
On-device gemstone classification
React Native ML Kit Edge AI

😊 RN-Sentiment-Analyser
Real-time sentiment analysis on mobile
React Native NLP TensorFlow Lite

🖥️ Activity Monitor ⭐3
Windows monitoring app • Screenshots every 5 min • Google Drive upload via OAuth 2.0
Python Google Drive API OAuth 2.0 System Monitoring

📊 Option Chain Analyser
NSE options data scraper • PCR & Volume PCR calculation • Real-time 5-min updates
Python Web Scraping Finance Data Analysis

🎓 Going Abroad
University data scraper • Auto-populates spreadsheet • Built during my own study-abroad search to replace manual Excel tracking
Python Web Scraping Automation

🤖 AtBSBI
MSc AI coursework repository covering:

• 🐄 Cow Pose Estimation (YOLOv8n-pose, ROS2 Jazzy, 57.5% mAP50 ↑, 59.7% size ↓)
• 🤖 Vision-Guided TurtleBot3 (ROS2 Humble, HSV tracking, LIDAR)
• 🌾 Crop Disease Classification (ResNet18, 92.3% accuracy)
• 🚁 Drone Delivery Framework (ROS2 Nav2, Pixhawk 6X, AprilTags)
• 💳 Credit Card Fraud Detection

PyTorch ROS2 Gazebo Computer Vision Deep Learning

🐍 Complete Python 3 Course ⭐20
Udemy Python course projects • Web scraping • Django • PyQt5 • Data viz • Speech recognition
Python Django PyMongo Web Scraping

🔥 Learn PyTorch
PyTorch framework experiments
PyTorch Deep Learning

🎓 Learning From Karpathy
Exercises from Andrej Karpathy's ML course • micrograd implementation
Python Neural Networks from Scratch

GitHub Achievements: 🌟 Starstruck • 🦈 Pull Shark • ❄️ Arctic Code Vault Contributor

📍 sanket-gadge.blogspot.com

• Dec 2025: How Seedr.cc Revolutionized My Torrent Downloading Experience
• Mar 2025: Understanding the German Tax System: A Student's Quick Guide
• Jun 2021: Training a Custom Named Entity Recogniser (spaCy)
• Feb 2021: Reinforcement Learning – UCB and Thompson Sampling

Topics: Machine Learning NLP Computer Vision Reinforcement Learning System Design

✅ DeepLearning.AI TensorFlow Developer (Coursera)
✅ Deep Learning Specialization (Coursera / DeepLearning.AI)
✅ Convolutional Neural Networks (Coursera)
✅ TensorFlow: Data and Deployment (Coursera)
✅ Improving Deep Neural Networks (Coursera)
✅ Data Science Math Skills (Duke University)
✅ Django REST API Development (Udemy)

Position: Werkstudent / Working Student
Focus Areas: Computer Vision • ML Engineering • MLOps • Robotics (ROS2)
Availability: 20 hours/week • Immediate start
Work Authorization: ✅ German residence permit (§16b) — No sponsorship needed

💼 Open to:

• Working Student positions in ML/CV/Robotics
• Research Assistant roles in Computer Vision
• Part-time MLOps/Backend (Python) opportunities
• Open-source collaborations

📧 Contact: gadgesanket75@gmail.com
📱 Phone: +49 176 76475359
🔗 LinkedIn: linkedin.com/in/sanket758
🌐 Portfolio: sanket758.github.io
📝 Blog: sanket-gadge.blogspot.com
🐦 Twitter: @sanket_gadge

• 🇮🇳 → 🇩🇪 From Mumbai to Berlin
• 🗣️ Trilingual: English (C1), German (A1, learning!), Marathi (Native)
• 📦 Published 2 PyPI packages: pan-aadhar-ocr & pdistrib
• 🎮 Built an AI that plays Snake autonomously (REINFORCE RL)
• 🤖 Working on vision-guided cow health monitoring with ROS2
• 📊 51 public repositories, 32 followers, still counting!

Last Updated: January 2025 | Built with ❤️ in Berlin