Building an end-to-end autonomous drone system that generates 3D Gaussian splat reconstructions. Built orbital flight generator with continuous POI tracking using DJI WPML/MSDK format. Implementing indoor positioning for GPS-free navigation and real-time object recognition.
Trained LSTM and Transformer models to automate tactic selection in the Lean theorem prover. Achieved 67% top-5 accuracy on the LeanDojo dataset (10K+ theorem-proof pairs). Targeting 40% reduction in manual proof construction effort.
Designed real-time anomaly detection combining YOLOv8 (object detection) with MobileLSTM (temporal action classification). 90% spatial accuracy and 97% temporal accuracy for RGB video-based public safety surveillance.
Drone-based system using YOLOv5s deep learning for automated tree counting and classification. Captured aerial images at 100m altitude, divided land into grids, and stitched images — achieving 95% detection accuracy across 500+ farmlands.
Classified 29 ASL alphabets + 3 special characters using Random Forest on 87,000+ images. Built NLP module for enhanced TTS output via Google TTS. Published: "ASL Recognition and Detection: A Review" at ICIVC (2023).
Sales analysis engine using the Apriori algorithm for cross-selling opportunity identification in large transactional databases. Extracted hidden patterns through data mining and generated association-based insights for marketing.
Real-time multi-user chat application with MongoDB-backed message storage and passing via Mongoose.