• Built and deployed a computer vision defect detection system for towel quality inspection (Python, OpenCV). It replaced a six-person manual process across 8-hour shifts, saving about ₹6,48,000 a month.
• Led setup, installation and commissioning of a CCCH Lockstitch Texpa Cross Hemming machine with German engineering experts.
• Owned daily preventive maintenance audits for CSP line machinery against OEM documentation.
• Got hands-on exposure to SAP in a live plant.
• Built all the Python and AI tooling myself in VS Code, since the corporate setup was limited to MS365.

Physics-Aware AI for Early Micro-Leak Prognosis in FESTO Pneumatic Systems (₹20,000 fellowship)

• Built a hybrid FluidSIM + real FESTO hardware setup with Siemens S7 PLC integration.
• Full ML pipeline: ingestion, preprocessing, feature engineering, and a VotingClassifier (Random Forest + Gradient Boosted Trees).
• F1 0.986 in simulation, R² 0.914 on hardware regression.
• Presented at TICSR 2026.

• Built and deployed a responsive web app: React frontend, Flask backend, REST APIs.
• Gathered requirements from non-technical stakeholders and delivered on time.
• Improved load speed and accessibility.

• Designed an automated quadcopter with IR sensors, Zigbee and embedded controllers for real-time environmental data collection (MATLAB, Python).
• Awarded best project of the ELC Summer Internship.

• AI Chatbot: Gradio + Hugging Face Transformers, memory use cut by 45% for CPU-only inference, live on Hugging Face Spaces
• Face Recognition Attendance: LBPH + OpenCV, 92% recognition accuracy
• Predictive Maintenance Classification: XGBoost, Random Forest, SVM, about 18% accuracy gain over an untuned baseline
• Sentiment Analysis: VADER on 10,000+ reviews, 91% accuracy
• Fake News Detection: BERT with PyTorch and Hugging Face, Flask app integrated with Azure
• Autonomous Buggy: IR + Zigbee transmitter/receiver boards
• Truss Design: fabricated truss that held a 6.8 kg load