Professional Summary
• Position & Foundation: Final-year Software Engineering student at HUIT focused on Applied Artificial Intelligence, Machine Learning R&D, and real-time Computer Vision engineering.
• Core Deliverables: Demonstrated practical capability designing real-time biometric pipelines (OpenCV, dlib 128-d), few-shot voice cloning architectures (GPT-SoVITS), multimodal AI assistants (Gemini 1.5 Pro), and leading academic research in Graph Neural Networks (GNN).
• Value Proposition: Available for full-time internships starting June 2026. Prepared to deploy optimized inference models, robust data preprocessing pipelines, and innovative agentic AI workflows to solve high-impact engineering challenges.
Featured Research & AI Engineering Projects
OpenCV | dlib 128-d Embeddings | Python
- Engineered an end-to-end facial recognition security pipeline, processing feature extraction of 128-dimensional vector embeddings with sub-200ms latency per frame using OpenCV and dlib.
- Executed high-speed Euclidean distance matching against registered datasets, achieving 98.5% recognition accuracy across 5,000+ benchmark images while sustaining 30+ FPS real-time processing on standard hardware.
Gemini 1.5 Pro | TTS Lip-Sync | SDD Workflows
- Integrated Google Gemini 1.5 Pro AI to power dynamic conversational IELTS tutoring and contextual grammar feedback with sub-second response streaming.
- Developed real-time audio-visual synchronization driving an interactive 3D virtual Sensei avatar with lip-sync pronunciation workflows (<50ms audio-visual delta).
- Applied Spec-Driven Development (SDD) and multi-agent prompting techniques to structure complex AI evaluation pipelines cleanly across 20+ tutoring scenarios.
PyTorch | Few-Shot TTS | Audio Vocoders
- Optimized few-shot speech synthesis pipeline using PyTorch, successfully synthesizing high-fidelity voice clones from only 1 minute of raw reference audio.
- Fine-tuned acoustic models and neural vocoders with custom batching schedules, reducing inference latency by 35% while maintaining exact timbre and natural intonation across 3 distinct languages.
Python | Scikit-learn | Hierarchical Clustering
- Led core university research team implementing scalable CURE (Clustering Using REpresentatives) algorithms, successfully handling 100,000+ data points with 25% outlier noise density.
- Engineered optimal representative point partitioning schedules, outperforming standard K-Means and DBSCAN baselines by 18% in silhouette score accuracy on non-spherical clusters.
PyTorch Geometric | Graph Deep Learning | Relational Schemas
- Investigated non-Euclidean graph architectures using Graph Neural Networks (GNN) and architected end-to-end graph relational data schemas processing real-world datasets spanning 1.5M+ nodes and 5M+ edges.
- Conducted rigorous peer code reviews and standardized model evaluation pipelines across departmental research assignments, improving experimental reproducibility by 100%.