OpenTyphoon.ai
Co-founded Thailand's open-source AI initiative as the only non-technical person on a team of 4. Led program management, GTM, partnerships, and operations from zero to 320k+ downloads.
320k+
model downloads
23M+
API calls
6k+
registered users
2k+
community members
what is typhoon
Typhoon is an open-source Thai large language model initiative, built under SCB 10X and positioned as Thailand's frontier AI research lab. The mission: advance open-source LLMs and multimodal technologies for the Thai language, with global-level research and local community engagement.
I co-founded this with a team of 4, three researchers and me. As the only non-technical person, I owned everything that wasn't model research: program management, go-to-market, partnerships, community, and operations. When I transitioned out, the team had scaled to 12.
program management
Typhoon wasn't one project. It was several running in parallel, and keeping them all moving without collisions was a big part of my job.
Data collection & annotation
Managed the pipeline for 3TB+ of public Thai language corpora and coordinated a data annotation team for instruction-following datasets. Collaborated with researchers on data approaches for supervised fine-tuning.
Model training coordination
While researchers handled the technical training, I managed timelines, resources, and dependencies across training runs, ensuring the team could move fast without stepping on each other.
Partnership portfolio
Tracked and managed relationships across research collaborations, academic partnerships, and API integration partners simultaneously.
Operations & admin
The unglamorous but essential work: team logistics, budgeting, tooling, and keeping a fast-growing research team running smoothly.
go-to-market
This is the part I'm most proud of. Building a great model means nothing if no one uses it. I treated Typhoon's open-source models the way a product team treats a freemium product: model-led growth, where the model is the product and developers are the users.
developer platform
Launched opentyphoon.ai with model access, API endpoints, and documentation, reducing the barrier from "clone a repo" to "hit an API."
community
Grew a Discord community from zero to 2,000+ members where developers shared use cases, reported issues, and built on Typhoon together.
technical content
Produced content showing real-world applications, including a YouTube feature on "The Rise of Intelligence" where I presented industry use cases like the customer support copilot built on Typhoon.
events
Organized and represented Typhoon at ACL 2024 in Bangkok (the premier NLP conference), Techsauce Global Summit (Southeast Asia's largest tech conference), and hosted a Typhoon Hackathon with 30 teams.
partnerships
Credibility in open-source AI comes from the company you keep. I built partnerships that gave Typhoon independent validation at a global level.
Stanford HAI
Collaborated with Professor Percy Liang's lab to build Thai language evaluation datasets and leaderboards on Stanford HELM, benchmarking Typhoon against 42 multilingual models for summarization and QA tasks.
AI Singapore (SEACrowd)
Joined the SEACrowd collaboration for Southeast Asian language data, positioning Typhoon within a broader regional research network.
Thai academia
Built collaborations with VISTEC, KMITL, and other Thai research and academic institutions, bridging industry application and academic research.
🎓The Stanford HELM benchmark gave us independent, globally respected evaluation. It wasn't just about the score. It was proof that a Thai-first model could compete on a world stage.
the numbers, in context
320k+
model downloads on Hugging Face
23M+
API calls through the platform
6k+
registered users
2k+
Discord community members
30
teams at the Typhoon Hackathon
4 → 12
team growth
a note on my role
I want to be upfront: I'm not a researcher. I didn't train the model or write the papers. My contribution was everything that turned Typhoon from a research project into something people actually used: the GTM, the partnerships, the community, the operations, the program management that kept multiple workstreams moving at once.
In a team of 4, there's no job description. You just do what needs doing. I'm proud that when I transitioned out, the foundation was strong enough for the team to keep growing.