A selection of my technical projects ranging across ML pipelines, robust backends, and cloud infrastructure.
Architected a full-stack containerized web application enabling sub-50ms semantic facial recognition for race participants. Implemented an asynchronous ingestion pipeline to decouple heavy DeepFace inference, achieving <100ms UI latency during bulk uploads.
Fine-tuned BERT and Llama-3 (using 4-bit QLoRA) for clinical NLP classification, achieving a 0.79 Macro-F1 score while reducing VRAM usage by 50%+. Orchestrated a robust ETL pipeline to process 53,000+ clinical text samples for supervised fine-tuning.
Built a real-time threat detection system coupling a Chrome Extension with a FastAPI backend to achieve <1s inference latency on live Twitter feeds. Stacked TF-IDF vectors with custom linguistic features to train a robust XGBoost classifier (90.7% accuracy).
Developed a dynamic Deep Learning classifier with Neural Architecture Search (Keras Tuner) to optimize network topology. Engineered multi-stage GridSearchCV pipelines for SVMs and integrated SHAP modules to translate raw tensor weights into interpretable health risk factors.
Designed a multimodal predictive regression pipeline fusing dense text embeddings with 100+ prosodic audio features. Used SentenceTransformers and Cosine Similarity to structure multi-turn dialogue, deploying K-Fold Cross-Validation to minimize prediction error.
Engineered a TrueNAS Scale home server cluster with a 24TB storage array. Orchestrated containerized microservices (Jellyfin, Immich, Tdarr) for high availability and seamless media transcoding, secured via Cloudflare Tunnels and Tailscale VPN.
Architected a secure, containerized full-stack marketplace application using Node.js and PostgreSQL. Implemented robust RESTful APIs with bcrypt encryption and asynchronous controller logic to aggregate persistent SQL user records with real-time session states.