AI researcher-builder working on language models, reasoning, post-training, evaluation, and agentic AI systems.

I study how language models behave, reason, fail, and improve — then turn those insights into open models, evaluation tools, datasets, applications, talks, and writing.

Currently, my work focuses on:

• 🧠 LLM reasoning & inference-time strategies
• 🧪 Post-training, SFT/RFT, RLHF, and behavior shaping
• 🧭 Evaluation, robustness, hallucination, and structured output following
• 🛠️ Agentic AI, context engineering, RAG, and tool-using workflows
• 🌏 Thai / Southeast Asian AI, open models, and low-resource language systems
• 🎨 Research × Engineering × Design — making ideas understandable, usable, and useful

I work on Typhoon, a family of Thai open models and applied AI systems, including reasoning models, medical reasoning, evaluation, OCR, ASR, and practical AI applications.

• Typhoon T1 — open Thai reasoning model research preview
• Typhoon-Si Med-Thinking 4B — medical reasoning model for ranked diagnoses

Explore more: Projects · Publications

My research asks questions like:

• Can we shape model behavior through post-training rather than only prompting?
• How do prior prompts affect reinforcement fine-tuning?
• When does hallucination hurt — and when can it unexpectedly help?
• How should we evaluate reasoning models beyond surface-level benchmark scores?
• How can we build more reliable AI systems for domains like medicine, finance, games, and Thai language tasks?

See: Google Scholar · ACL Anthology

I enjoy building tools that help people evaluate, understand, and apply AI systems.

Some public projects include:

• 🧪 BenchING — structured output benchmark for LLMs
• ⚖️ Themis — lightweight platform for LLM evaluation experiments
• 🏷️ Thoth — GUI tool for dataset labeling
• 🚗 DriveSSD / ADAS distillation — cost-effective LLM distillation for driving assessment
• 🇹🇭 Krathu-500 — Thai post-comment corpus
• 🎮 LLM story generation evaluation — studying LLMs in narrative and game contexts

Browse more: github.com/Pittawat2542

I write and speak about AI research, LLM reasoning, post-training, agentic workflows, and the future of AI systems.

Recent topics include:

• Agentic AI with Context Engineering
• Open Models, Smarter Agents
• How to Train Your Reasoning Models
• From Text to Thought: How Modern AI Models Learn to Reason
• Open-Source Typhoon: Democratizing Advanced AI
• AI as the Common Language of Knowledge

Read more: Blog · Talks

I’m interested in AI systems that are not only powerful, but also:

• reliable enough to evaluate honestly,
• useful enough to solve real problems,
• open enough to expand access,
• and understandable enough for people to reason with.

My long-term goal is to help push the boundary of knowledge while building AI that meaningfully serves humanity.

• 🌐 Website: petepittawat.dev
• 📚 Publications: Google Scholar
• 💻 GitHub: Pittawat2542
• 📧 Email: work@petepittawat.dev