Asaad Shaikh

Engineered a low-latency local speech-to-speech conversational pipeline orchestrating STT, a quantized LLM, and TTS across dual RTX 3060 GPUs. Achieved 600–1200ms end-to-end voice response latency optimizing model loading and inter-stage data flow for real-time conversational interaction.

Built a full-stack semantic search app using Google Gemini embeddings and Supabase pgvector to match user preferences to movies via vector similarity. Integrated a RAG pipeline where retrieved results are formatted by Gemini into structured JSON responses, served through a Node.js/Express REST API with a dynamic vanilla JS frontend.

Built a full-stack translation web app with client-server architecture powered by OpenAI API, featuring a Node.js/Express backend API, dynamic frontend rendering with Markdown parsing and XSS sanitization, and real-time user feedback with loading states.

Built a full-stack conversational AI chatbot using Google Gemini and Node.js/Express, featuring multi-turn dialogue, context management, intent-aware responses, and a RAG pipeline backed by Supabase pgvector for knowledge base retrieval.

Binary and multiclass classification with Random Forest, XGBoost, and neural networks on the UNSW-NB15 dataset. Performed data pre-processing, class balancing, feature engineering, PCA, hyperparameter tuning, cross validation, and experimented with various SMOTE techniques. Analyzed multiple existing research studies to inform the approach.

Engineered and tuned a gradient boosting classifier using XGBoost with hyperparameter optimization via GridSearchCV, achieving 92% accuracy in identifying students at risk of dropping out. Included data engineering, cleaning, and evaluation using confusion matrix and classification metrics.

Built a workflow automation system for a real university client, featuring role-based access control, digital approval signing and tracking, automated notifications and reminders, and a fully responsive interface. Developed using integrated Agile test-driven development practices.

Multi-page static site for a fictional semiconductor foundry (Kamloops Semiconductor Manufacturing Company). Built with semantic HTML, modular CSS, and vanilla JavaScript—product pages, contact and order forms with client-side validation, and a responsive multi-section layout.

Regression project predicting in-game FPS from CPU and GPU hardware features. Full pipeline: EDA, missing data and outlier analysis, correlation and VIF, PCA, and model comparison with Random Forest, Gradient Boosting, AdaBoost, and MLP using GridSearchCV and cross-validation.

Flask application for COMP 3260 demonstrating practical security controls: Fernet encryption for sensitive fields, RSA-PSS signatures on uploaded documents, SQLite persistence, and an IP firewall with logging, blocking, and unblock workflows.

Flask demo that runs BB84 quantum key distribution on real IBM Quantum hardware via Qiskit Runtime, derives a symmetric Fernet key from the sifted bit string, and exposes a minimal encrypted messaging UI to illustrate QKD-backed symmetric encryption.