Thein Kyaw Lwin

Thein Kyaw Lwin is an ML/AI researcher and data scientist specialising in computer vision, natural language processing (NLP), efficient deep learning, and edge AI. He is currently pursuing a Doctor of Philosophy in Engineering Education at Batangas State University, while serving as a Research Assistant at BatStateU and contributing as a Researcher at MMDT-RISE. He holds a Master of Science in Data Science (GWA: 1.1250, highest is 1.0) from BatStateU as a fully funded IIE Myanmar Student Emergency Support (MSES) Fund Grantee.

He combines academic research with practical engineering to develop scalable AI solutions for real-world deployment. Supported by seven years of field technology experience across agriculture, public health, and enterprise IT, his research focuses on practical, scalable AI systems that bridge methodological innovation with real-world impact.

Selected proof points

Edge AI & Computer Vision: Designed a hybrid lightweight CNN (SqueezeNet v1.1 + CBAM) for chickpea disease detection, achieving 82.09% cross-domain accuracy at 0.870 MB (3.31× compression via INT8 quantization), validated across 115 real Android devices via Google AI Edge Portal; published in Procedia Computer Science (ICCSCI 2026).
Low-Resource NLP & Transformers: Co-authored research on Burmese grammar-driven text segmentation (Language Resources and Evaluation, Springer — under review), Burmese Named Entity Recognition (ACM TALLIP — under review), and text segmentation representation quality (ICLR 2027 — submitted).
Generative AI & LLM Systems: Built and deployed Shwe Taungthuu, a Gemini 2.5 Flash bilingual agricultural advisory agent on Google Cloud Run with Burmese voice/text input (presented at Google Developers Community Build with AI MMDT 2026).
Geospatial NLP: Built A2C Myanmar Address-to-Coordinates, fine-tuning multilingual transformers on 610,509 Myanmar addresses with a custom Haversine distance loss and a live Streamlit demo.
Health Systems & Infrastructure: Administered DHIS2 and OpenMRS eHealth infrastructure at WHO Myanmar for 1,000+ users with 99.99% uptime.
ICT4D & Agriculture: Built agricultural data pipelines at CESVI Myanmar reaching 130 villages and 26,234 households, with mobile application training for 2,000+ farmers.

Featured work

Chickpea Fusarium Wilt Detector — offline Android edge-AI plant disease detection for smallholder farmers.
Shwe Taungthuu — bilingual Gemini agricultural advisory agent on Google Cloud Run.
A2C Myanmar Address-to-Coordinates — transformer-based geospatial NLP for Myanmar address geocoding.
Bean Disease Classification — stepwise CNN fine-tuning achieving 98.86% test accuracy.
Wikipedia Web Traffic Prediction — comparative time-series forecasting study across 145,000 articles.
ML Algorithms from Scratch — classical ML implementations in NumPy with first-principle derivations.
MMDT Deep Learning Projects — applied deep-learning coursework spanning tabular regression, NLP, and CNN benchmarking.

Core stack

PyTorch · TensorFlow/Keras · LiteRT · Vertex AI (Gemini 2.5 Flash) · Google Cloud Run · FastAPI · Python · R · SQL · QGIS · Docker · Linux · Android/Kotlin · Weights & Biases

For detailed background, see the CV, publications, project portfolio, GitHub, or LinkedIn.

selected publications

ICCSCI 2026

A Systematic Ablation of CBAM Depth in SqueezeNet for Optimising Cross-Domain Chickpea Disease Detection

Thein Kyaw Lwin, Rowell M. Hernandez, Princess Marie B. Melo, and 2 more authors

Procedia Computer Science, 2026

11th International Conference on Computer Science and Computational Intelligence (ICCSCI 2026)

DOI Bib HTML

@article{lwin2026systematic,
  title = {A Systematic Ablation of CBAM Depth in SqueezeNet for Optimising Cross-Domain Chickpea Disease Detection},
  author = {Lwin, Thein Kyaw and Hernandez, Rowell M. and Melo, Princess Marie B. and De Chavez, Poul Isaac C. and Macatangay, Lloyd H.},
  journal = {Procedia Computer Science},
  volume = {289},
  pages = {1259--1266},
  year = {2026},
  publisher = {Elsevier},
  doi = {10.1016/j.procs.2026.08.346},
  note = {11th International Conference on Computer Science and Computational Intelligence (ICCSCI 2026)},
}

Springer

Grammar-Driven Text Segmentation for Context Understanding of Myanmar Language

Myo Thida, Nu Wei Thet, and Thein Kyaw Lwin

Language Resources and Evaluation, 2025

Currently under peer review (contributed to Testing). Preprint Version 1

DOI Bib HTML

@article{thida2025grammar,
  title = {Grammar-Driven Text Segmentation for Context Understanding of Myanmar Language},
  author = {Thida, Myo and Thet, Nu Wei and Lwin, Thein Kyaw},
  journal = {Language Resources and Evaluation},
  year = {2025},
  publisher = {Springer},
  doi = {10.21203/rs.3.rs-8408602/v1},
  note = {Currently under peer review (contributed to Testing). Preprint Version 1},
}

ACM TALLIP

Burmese Named Entity Recognition: Annotated Corpus and Multilingual Transformer Benchmark

Myo Thida, Nu Wai Thet, Khine Zar Thwe, and 1 more author

ACM Transactions on Asian and Low-Resource Language Information Processing, 2026

Currently under peer review

Bib

@article{thida2026burmese,
  title = {Burmese Named Entity Recognition: Annotated Corpus and Multilingual Transformer Benchmark},
  author = {Thida, Myo and Thet, Nu Wai and Thwe, Khine Zar and Lwin, Thein Kyaw},
  journal = {ACM Transactions on Asian and Low-Resource Language Information Processing},
  year = {2026},
  publisher = {ACM},
  note = {Currently under peer review},
}

ICLR 2027

Evaluating Text Segmentation for Representation Quality in a Low-Resource Language

Thein Kyaw Lwin, Myo Thida, Nu Wai Thet, and 1 more author

In International Conference on Learning Representations (ICLR 2027), 2027

Submitted

Bib

@inproceedings{lwin2027evaluating,
  title = {Evaluating Text Segmentation for Representation Quality in a Low-Resource Language},
  author = {Lwin, Thein Kyaw and Thida, Myo and Thet, Nu Wai and Thwe, Khine Zar},
  booktitle = {International Conference on Learning Representations (ICLR 2027)},
  year = {2027},
  note = {Submitted},
}