Thadoe Hein
Bridging research and production systems, specializing in MLOps, LLM systems, and scalable AI platforms.
About Me
AI Engineer with strong experience bridging research and production systems. I specialize in designing end-to-end machine learning pipelines, production-grade RAG systems, and multi-agent workflows.
Experienced in model lifecycle management, CI/CD pipelines, and deploying AI services with monitoring, logging, and performance optimization.
AI & ML
Frameworks
Programming
Infrastructure
Experience
AI Developer (Freelance)
SoeMindAI, Inc.
- Cut AI costs by ~30% via optimized prompt design and model selection
- Automated end-to-end presentation slide creation, reducing multi-hour workflow to minutes
- Improved unified study platform with 10 learning tools including flashcards and mind maps
- Engineered job prep platform with ATS compatibility and interview prep
Machine Learning Engineer Intern
Language Understanding Laboratory (LU Lab)
- Fine-tuned transformer models (mT5) to improve EM/F1 scores
- Built reproducible ML pipelines using Hugging Face
- Conducted quantitative research for model optimization and validation
AI Engineer Apprenticeship
Intelligent Systems Laboratory (ISL)
- Designed and deployed production-grade RAG systems
- Built end-to-end ML pipelines including data ingestion and model serving
- Operationalized LLM applications via REST APIs
- Implemented monitoring for model performance and latency
Projects
Video Intelligence & Face Search Platform
Search people in videos using face similarity and embedding-based retrieval
- •90% reduction in video processing time
- •Sub-second identity search across large datasets
- •Scalable AWS architecture (Lambda, ECS, RDS)
RAG Platform for Document Intelligence
Document-based chatbot using retrieval-augmented generation
- •Improved answer accuracy through document retrieval
- •Reduced response latency via automated pipelines
- •Enhanced quality with ranking strategies
End-to-End OCR & Vision Pipeline
Computer vision pipeline for receipt text detection and localization
- •50+ hours saved via automated annotation
- •0.93 text detection accuracy on Thai receipts
- •Custom YOLOv11 fine-tuning
LLM Fine Tuning for News Summarization
Fine-tuned FLAN T5 model for improved news summarization
- •Improved summarization metrics
- •Reproducible training pipeline
- •Systematic hyperparameter tuning
Get in Touch
I'm always interested in hearing about new opportunities, collaborations, or just having a chat about AI and technology. Feel free to reach out!
Get to know me beyond work.
A little more about who I am outside work - the places I love, the games I enjoy, and the moments I collect while traveling.
© 2026 Thadoe Hein.
Want to know me beyond projects and work experience?
Meet Me Beyond Work