TL;DR
Typhoon is Thailand's first open-source multimodal AI initiative, including LLMs, vision, and voice models built for the Thai language. As a founding member, I led program management, GTM, partnerships, and community, scaling from zero to 320k+ model downloads, 23M+ API calls, and 6k+ developers on the platform.
- Role
- Founding Product/Program Manager
- Timeline
- May 2023 – Mar 2025
- Tech Stack
- PyTorch · Hugging Face · vLLM
- Skills
- Technical Program Management, GTM, Partnerships, Community
The Problem
Thai is one of the world's most underrepresented languages in AI. Most large language models are built for English first, and when they do support Thai, they often miss context, culture, and how people actually communicate.
Ask ChatGPT in Thai how to cook grilled chicken (ไก่ย่าง) and it'll give you a fluent Thai recipe that tells you to put the chicken in the oven. In Thailand, ไก่ย่าง means chicken skewers over charcoal. Most Thai kitchens don't even have ovens.
The language was right but the context was wrong, so we built Typhoon models to narrow this gap.
The Solution
Building open-source multimodal AI models for Thai language and culture, so anyone can start using them for Thai-specific use cases.
User picks the right model
Typhoon is a family of models across every modality: text, speech, vision, and translation. Users browse the collection, pick the model that fits, and can deploy it locally.
User tries the model directly in the browser
Before committing to anything, anyone can test any Typhoon model at playground.opentyphoon.ai. No extra setup needed.
User builds production-grade applications with API
Full API access with documentation and demo apps. Same models as the playground, ready for production.
Research & Partnership
Typhoon: Thai Large Language Models
Dec 2023Co-authored technical report on the development of Thailand's first open-source Thai LLM, covering data preparation, pretraining, instruction-tuning, and evaluation. Typhoon achieved performance on par with GPT-3.5 in Thai while being 2.62× more efficient at tokenizing Thai text.
Kunat Pipatanakul, Phatrasek Jirabovonvisut, Potsawee Manakul, Sittipong Sripaisarnmongkol, Ruangsak Patomwong, Pathomporn Chokchainant, Kasima Tharnpipitchai
arxiv.org/abs/2312.13951 (70+ citations) →
Worked with Professor Percy Liang's lab at Stanford HAI to add Thai to the HELM leaderboard, benchmarking Typhoon against 42 multilingual models.
crfm.stanford.edu →
Joined AI Singapore's SEACrowd collaboration to pool Southeast Asian language data, placing Typhoon inside a broader regional research network.
seacrowd.org →
Ongoing research partnership with one of Thailand's leading science and technology institutes, spanning multiple projects across Thai NLP.
vistec.ac.th →Other partners and collaborators: SambaNova, TogetherAI, Mahidol University, Thammasat University, KMITL, InnovestX.
Community Events
