Software development, machine learning, web applications, and everything in between. These projects represent the digital side of my work.
Modeled internal transformer representations as trajectory dynamics in high-dimensional state space. Demonstrated representations lie on low-dimensional manifold, enabling long-horizon forecasting with low-rank state-space predictors, reducing compute by 22-50%. Developed orthogonal projection framework for behavioral steering.
Developed a Single Threaded Web server in Rust parsing HTTP requests. Extended to multi-threaded server. Implemented a Distributed File System based on Sun Microsystems’ NFS enabling remote file access and seamless data sharing. Integrated an RPC layer for low-latency client-server communication.
Implemented zero-reference deep curve estimation for frame-wise video light enhancement. Retrained YOLOv7 for real-time face detection and DeepSort for multi-object tracking. Applied GANs for landmark detection, frontalization, and reconstructed latent facial details using GFG-GAN restoring high-fidelity facial textures.
Developed cross-platform app allowing facial recognition for attendance, geolocation recording, events management. Implemented Face Detection using HOG and SVM, and Anti-Spoofing using Gesture Detection via MediaPipe. Integrated containerized models within a Flutter app using Flask and Docker.
Developed a robust RAG application based on dynamic retrieval strategies and multi-agent orchestration for analytics and compliance. Implemented Graph-Based Text Indexing and MCTS/Markov Chain algorithms to optimize workflows. Prototyped advanced frameworks like MetaGPT, AFlow, and Hybrid RAG.
Developed a production-grade, modular and autonomous Agents framework orchestrating OAuth flows for GitHub, Jira, Gmail, etc. Implemented stateful session management via Rclone and dynamic agentic workflows. Architected scalable ETL pipelines and leveraged Milvus and mem0 for advanced vector search and long-term memory.