Võ Duy Hùng

Co-authored research on speech processing for a low-resource Bahnaric language and implemented custom TTS and voice-conversion models, resulting in publications accepted to VNUHCM Journal of Engineering and Technology.

Designed end-to-end data-cleaning pipelines and developed a hybrid classification + similarity-search system for math problems, improving retrieval robustness; led R&D on a multi-persona chatbot and prototyped persona-specific dialogue features.

Fine-tuned Stable Diffusion models with LoRA for virtual model generation, developed a proprietary virtual-try-on pipeline and optimized inference using TensorRT and Triton to reduce latency and increase throughput for Modeli.ai.

Enhanced HR chatbot modules using QLoRA quantization and knowledge distillation, built a code-generation and knowledge-transfer chatbot with RAG and chain-of-thought, and integrated services with Rancher, FastAPI, and Docker.

Researched and implemented hybrid RoBERTa+LSTM models for text-to-emotion and an end-to-end auto-question-generation pipeline; improved text-relevancy using embedding cosine similarity and deployed services on AWS with FastAPI and Docker.

Master of Science in Signal Processing and Machine Learning at Tampere University, expected May 2027.

Activities and societies: Thesis: VIEACT-TTS-VC; IELTS: 7.0

Bachelor of Engineering in Computer Science with a thesis on a Vietnamese end-to-end TTS and voice-conversion framework accepted to the 10th ASIA International Conference 2024.

Completed high school specialization in Physics at VNU-HCM High School for the Gifted, graduated May 2020.