Live timing · Bangkok · updated Sept 2026

Papatchol
“Sand” Thientong

PhD student in Linguistics at Chulalongkorn University, studying why LLMs sound cold in Thai.

Team radio

“Parsed 200 million Thai dependency arcs. Still can’t parse my own to-do list.”

Read ThaiTrees Download CV Say hi
Pos Entry Status
1 PhD LinguisticsChulalongkorn University · Year 1 In session
2 ThaiTreesarXiv:2609.27558 · 342M tokens Personal best
3 Eccelix AI SolutionsCo-founder Running
4 MSc Speech & Language ProcessingUniversity of Sheffield Distinction
active personal best finished

P1 Now

Portrait of Sand Thientong
#23 Papatchol Thientong Chula Linguistics · Year 1
  • Studying Year 1 of a PhD in Linguistics at Chulalongkorn University, supervised by Assoc. Prof. Attapol Thamrongrattanarit-Rutherford. This semester: NLP III, Psycholinguistics, and Phonology & Syntax.
  • Building Eccelix AI Solutions, which I co-founded with four friends from Sheffield.
  • Just shipped ThaiTrees, a 342-million-token parsed corpus of Thai across four domains, now on arXiv.
  • Off track Six-hour team endurance races in the EasyKart Championship, cooking for my two dogs, and studying poker theory I don’t use.

P2 Research

Dissertation direction

Why does AI-generated Thai sound polite but cold?

English builds empathy mostly from emotion vocabulary. Thai carries much of it in grammar: sentence-final particles like นะ and เนอะ, set expressions like ไม่เป็นไร and สู้ ๆ, and leaving things unsaid. My hypothesis is that LLM Thai gets the grammar and politeness right but misses these interactional signals. I’m testing this in everyday comforting conversations.

  • Thai treebanks
  • Dependency parsing
  • Register variation
  • Affect & empathy in LLMs

Publication

Rutherford, A. T. & Thientong, P. (2026). ThaiTrees: Thai Syntactic Distribution Across Domains. arXiv:2609.27558.

342M tokens 4 domains 199.8M dependency edges 451k word forms
Paper Corpus & data

P3 Pit lane

Education and work, most recent first.

  1. 2026–now

    PhD in Linguistics

    Chulalongkorn University, Bangkok

    Thai NLP and empathy in AI-generated Thai, in Prof. Attapol’s lab. A grades in Computational Linguistics (NLP II) and Semantics, transferred into the programme.

  2. 2025–now

    Co-founder

    Eccelix AI Solutions

    An AI solutions company started with four friends from Sheffield.

  3. 2024–25

    MSc Computer Science with Speech and Language Processing

    University of Sheffield, UK

    Distinction (weighted average 75.83; dissertation 77). Dissertation: Thaislate, an LLM-based Thai–English grammar learning tool.

  4. 2022–23

    Chief Operating Officer

    Interpal English School, Bangkok

    Built the school’s operational framework and its “Hybrid English” teaching method.

  5. 2019–23

    BEng Automotive Engineering

    Chulalongkorn University International School of Engineering

    Class president and CFO of the Chulalongkorn University Racing Electric club. This is where the karting started.

  6. 2018–22

    English Tutor & Team Leader

    Monkey Monkey Tutoring School, Bangkok

    Taught classes of 1 to 30 students and led the part-time tutor team. Also taught a TOEIC course at Valaya Alongkorn Rajabhat University in 2023.

IELTS 8.5TOEIC 985/990Thai (native) · English (C2)Python · PyTorch · Stanza · PyThaiNLP · LangGraph · XeLaTeX

P4 Garage

Corpus · 2026

ThaiTrees

A parsed corpus for comparing Thai syntax across news, Wikipedia, speech and social media. The pipeline runs CRFcut, AttaCut, AttaParse and a PhayaThaiBERT POS tagger I fine-tuned for the paper. The domains split into a formal pair (news and Wikipedia) and a conversational pair (speech and social media), by word frequency and by syntax.

90.64% POS accuracy 4.5M sentences 5,944 edge patterns
Paper GitHub

MSc dissertation · 2025

Thaislate

A translation tool that explains English tenses to Thai learners in Thai. Typhoon-Translate-4B translates, an XLM-RoBERTa classifier identifies the tense, and Typhoon-2.1-Instruct writes the explanation. It ran as a Flask app on GCP and was tested in an ethics-approved user study.

94.7% tense accuracy 38 participants 4.33/5 rating
GitHub Dissertation Short version Blog series

Earlier work

  • Traffic signal optimisation: an AnyLogic simulation of a Bangkok intersection that improved traffic flow by 27%.
  • Water pH analysis: stepwise linear regression of drinking-water pH against weather factors (R² = 0.36).

P5 Blog

The Thaislate series (2025): five posts on building my MSc dissertation, from the idea to the user study.

  1. 01Introducing Thaislate: AI-powered grammar learning8 min
  2. 02Building the neural pipeline: from translation to explanation6 min
  3. 03Training a custom BERT classifier for Thai–English tenses7 min
  4. 04User experience design for language learning5 min
  5. 05Measuring success: user analytics and performance6 min

P6 Radio

Part-time introvert, full-time overthinker. I went from teaching people English to teaching machines Thai, and now I’m working out why the machines still sound like a customer-service script. Most race weekends I’m at the kart track pretending to be Alex Albon (fellow Thai representing!) and planning endurance stints to three decimal places before driving them on instinct anyway. Off track, I cook home-made meals for Afong and Otueng (my dogs eat better than I do), read GTO poker charts instead of winning actual poker, and play Valorant and FC Online with a Man Utd historical squad.

papatchol.t@gmail.com linkedin.com/in/papatcholt github.com/sandpapat

Line ID: dewsand