PROJECTS

Python, OpenCV, Arduino (C), Raspberry Pi

• Developed a real-time color detection pipeline in Python using OpenCV on Raspberry Pi for object classification

• Implemented multi-servo robotic control in Arduino (C) to actuate base, arm, and gripper for automated sorting

• Designed a distributed control architecture with Raspberry Pi handling vision processing and Arduino handling motion
control

• Integrated serial communication between controllers to achieve synchronized detection-to-actuation workflow

• Automated boot-time execution and deployment scripts to enable fully autonomous system startup

Python, PyTorch, YOLOv8, OpenCV, Grad-CAM, Google Colab

• Developed an automated PCB defect detection pipeline using YOLOv8 for accurate identification of manufacturing faults

• Utilized the HRIPCB dataset and performed preprocessing to prepare images and annotations for YOLOv8 training

• Fine-tuned the model over extended training runs (≈12 hours on GPU) improving accuracy from 92.5% to 96.8% and recall
from 93.6% to 99.4%

• Achieved strong detection performance with 98.4% F1-score and 98.7% mAP@50 through hyperparameter tuning and
augmentation strategies

• Implemented Grad-CAM visualizations to generate interpretable heatmaps highlighting defect-relevant regions

Next.js, Supabase, DeepSeek API

• Designed and implemented a structured Manchu–Latin–English database using Supabase to store and manage multilingual
lexical entries

• Digitized and curated historical dictionary sources, processing and validating hundreds of entries for consistency

• Built a responsive Next.js interface with search, gallery, and translation views supporting paginated browsing and filtering

• Managed archival image storage and retrieval using Supabase Buckets with optimized linking to database records

• Integrated DeepSeek API to provide on-demand Manchu–English translation within the web interface

ARNAV PANWAR