Ho Tan Dat
AI Researcher & PhD Candidate
Download CVAI Researcher and PhD candidate at the University of Debrecen on a fully funded Hungarian State scholarship. Building production AI systems from LLM inference and fine-tuning to multi-stage NLP pipelines, computer vision, and time-series forecasting. Bridging academic research with hands-on engineering across microservices architecture and full-stack development.
- Built a multi-service compute and AI inference platform with microservices architecture (FastAPI, Nginx reverse proxy, Redis, MongoDB), serving as a unified research computing workspace
- Designed and deployed an LLM inference service powered by vLLM with OpenAI-compatible API, supporting streaming chat completions, embeddings, and dynamic model load/unload management
- Developed an AI text humanizer with a 3-stage NLP pipeline (paraphrasing, word perturbation, grammar correction) using a LoRA fine-tuned Qwen3-32B model, with parallel chunk processing and async job queue
- Implemented a document extraction pipeline using Vision Language Models (VLM) to extract structured data from uploaded PDFs via asynchronous task workers
- Developed time-series forecasting models (Temporal Fusion Transformer, DeepAR) for sales predictive analytics with weather and holiday effect integration
- Designed a centralized API gateway with JWT authentication, per-service API key injection, and unified OpenAPI documentation aggregation
- Produced research papers for publication in international conferences and journals
- Leveraging AI models (Transformers, BERT) to extract structured data from unstructured sources
- Labeling and training custom AI models for task-specific resolutions
- Developing modules that integrate LLM APIs for business requirements
- Building RESTful APIs with FastAPI for scalable application architectures
- Implemented embedding-based RAG solution combining semantic retrieval with generative AI
- Integrated AI models into client systems (face recognition, vehicle plate recognition)
- Developed RESTful APIs using FastAPI to integrate AI into web applications
- Containerization, Kubernetes (K8s), and Google Cloud Platform (GCP) deployments
- Established and validated hypotheses on training datasets for robust model performance
- Participated in code reviews, sharing Python development best practices
- Assisted faculty in delivering Software Project Management, Software Testing, and Basic Programming courses
- Facilitated lab sessions and provided hands-on support with coding assignments
- Created and delivered course content aligned with current industry practices
- Mentored students through academic journey, project work, and thesis development
- Conducted research in Machine Learning focusing on algorithm development and applications
- Published research findings and presented at conferences
Deployed and managed vLLM inference servers with OpenAI-compatible APIs, streaming completions, and dynamic model management. Fine-tuned large language models (Qwen3-32B) using LoRA adapters for domain-specific tasks
Built multi-stage NLP pipelines (paraphrasing, perturbation, grammar correction) with concurrent processing. Experienced with prompt engineering, embedding-based RAG solutions combining semantic retrieval with generative AI
Applied vision models for face recognition, vehicle plate recognition, and object detection (YOLO, Grounded-SAM). Built document extraction pipelines using Vision Language Models to parse structured data from PDFs
Developed production forecasting models using Temporal Fusion Transformer and DeepAR for predictive analytics with weather and holiday effect integration across grouped time-series data
Designed and built multi-service platforms with API gateway routing, JWT authentication, per-service key injection, async task queues (Redis/RQ), and unified OpenAPI documentation aggregation
Managed multi-GPU server environments for AI inference, built real-time monitoring dashboards, sandboxed code execution with GPU access, and process management via Supervisor/systemd
Containerization with Docker, orchestration with Kubernetes, and cloud deployments on GCP. Experienced with Nginx reverse proxy configuration, CI/CD workflows, and production service management
Published and submitted papers in international conferences and journals (Springer, IGI Global). Skilled in identifying research gaps, synthesizing concepts, and translating findings into practical applications
End-to-end application development from vanilla JS frontends to Python backends with MongoDB/Redis data layers. Deep understanding of the full software development life cycle