Typhoon / 2022–2025 / Visit here (opens in a new tab)
Patt Chokchainant.
I have six years of experience as a program manager turning AI research into products at scale.
I led program operations at Typhoon, an open-source Thai LLM, where I managed our data annotation pipeline and launched a multilingual model evaluation in collaboration with Stanford HAI.
Working with researchers, engineers, and founders every day inspired me to deepen my technical skills, so I came to the University of Washington to pursue my master’s degree.
Outside of work, I enjoy hiking, reading, playing boardgames, and traveling. Feel free to send any recommendations!
Previously at:
- Coffey venturesProduct2026Restaurant OS for independent coffee shops (20K+ merchants)
- The AI Collective, Seattle ChapterEvent & BizOps2026Global AI community of 100K+ founders, researchers, operators and investors
- SCB 10X AI LabProgram & Strategy2025Typhoon (founding member): open-source Thai LLMs (23M+ API calls, 8K+ users)Data collection: 5K financial Q&A pairs via annotation vendor managementEnterprise deployment: AI copilot for financial services customer service
- Speedboat StudioCo-founder2023No-code NFT creator platform (18K+ global users, $4M sales volume)
- FinnomenaProduct2022Wealth management platform ($1B+ AUM), platform for financial advisors
- SertisTech Consulting & Delivery2021Data and AI consulting firm, 7-Eleven shopping chatbot (16M+ MAU)
Some fun stuff I did:
- Hiked 48 miles with 6,660 elevation gain in 5 days
- Ran my first 10K in Feb 2026 in 55:56 minutes
- Been to 24 countries across 4 continents
Projects
The AI Collective, Seattle Chapter (opens in a new tab) / 2026 / Visit here (opens in a new tab)
PromptHub (opens in a new tab) / 2023 / Product Hunt (opens in a new tab) / Demo (opens in a new tab)
SnapStrip (opens in a new tab) / 2025 / Try here (opens in a new tab)
Speedboat Studio (opens in a new tab) / 2023 / No longer active / Visit here (opens in a new tab)
Finnomena / 2021–2022 / Visit here (opens in a new tab)
7-Eleven AI shopping chatbot / 2019–2021
Typhoon
Thailand’s first open-source LLM
Typhoon is the first Thai AI model family. As a founding member, I led program management, GTM, partnerships, and community, scaling to 320k+ model downloads, 23M+ API calls, and 8K+ developers on the platform.
opentyphoon.ai ↗arXiv 2312.13951 ↗
- Role
- Senior Program Manager · 2023-2025
- Industry
- AI, Open-source LLMs
- Tech Stack / Tools
- Annotation Platform (Vendor Portal) · Hugging Face · SQL · Google Sheet · Notion · Google Analytics · Discord
- Skills
- Program Management · Data Annotation · Vendor Management · Go-to-Market · Community Building · Partnerships · Enterprise Deployment
Problem
Thai is a low-resource language as it lacks sufficient training data to build NLP systems. Additionally, western models don’t understand Thai language and culture.
Solution
Build accessible open-weight models and distribute them to local Thai developer communities.
1. Open-weight models
The Typhoon model family is available on Hugging Face, ready to download and run locally.
2. Production-grade APIs
Hosted APIs with world-class documentation.
3. Zero-setup developer playground
Anyone can try any Typhoon model in the browser at the developer playground.
I managed four work streams:
Model Training & Data Collection
- Managed the data collection program end-to-end through annotation vendor management and quality review, delivering 5K+ financial Q&A pairs
- Worked with AI researchers on 3TB+ pretraining data curation
- Co-authored the technical report
Events & Community
Research Partnerships
- Stanford HAI (Professor Percy Liang's lab): a Thai evaluation dataset on the Stanford HELM leaderboard
- SambaNova: Typhoon 8B and 7B models on Samba-1 CoE model
- AI Singapore: the SEACrowd project for Southeast Asian language data
- Guardrails AI: a Thai language safety dataset
- VISTEC: Thai language data collection projects
Enterprise Adoption
- Delivered a RAG-based AI copilot for customer service at Thailand's largest bank. View more here
AI Copilot for Customer Support
A RAG-based AI copilot for financial advisory support
SCB 10X is the innovation arm of SCBX, the parent group of Siam Commercial Bank, Thailand’s largest bank by market cap at $10B+. Its financial advisory customer service team handled 15,000 tickets a day (~100 per day per agent), with predominantly manual processes.
I delivered a RAG-based AI copilot to production with AI engineers and the client team. It cut information retrieval time for 150+ agents from days to under 5 minutes.
- Role
- Senior Technical Program Manager · 2023-2024
- Industry
- Financial Services / Financial Advisory
- Tech Stack
- LangChain · Pinecone · LangFuse · Salesforce CRM and Knowledge Base
- Skills
- Product Management · User Research · Evaluation
Problem
One ticket could take hours, or days if it is a cross-department question.
- Scattered knowledge: Information lived across PDFs, the knowledge base, Salesforce CRM, and shared drives, all with inconsistent formats.
- Manual ticket summaries: Agents wrote every summary by hand after each chat. Information sometimes got lost, was inaccurate, or incomplete.
- Strict financial regulations: Every design choice had to comply with SEC Thailand regulations and data privacy policy.
Solution
A RAG-based AI copilot designed for customer service agents to streamline information retrieval and time to resolution with human oversight.
- Real-time knowledge retrieval: Semantic search across Salesforce knowledge (policies, troubleshooting guides, regulatory docs) and customer data. Sources appear next to each answer so agents can review them.
- Reply suggestions: The copilot drafts three replies per message, each grounded in retrieved sources. Agents pick one, edit, and send.
- Auto-generated ticket summaries: The system writes a summary and next steps back to Salesforce after each conversation.
System architecture
PDPA-compliant PII redaction: Sensitive customer data was masked before routing to the LLM, then re-injected back into the final output.
Model evaluation: Automated retrieval tests and LLM-as-judge scoring against a golden dataset built with support leads, plus LangFuse tracing and agent feedback to improve each iteration.
Result: Information retrieval time dropped from days to under 5 minutes for 150+ agents.
Speedboat Studio
TL;DR
[placeholder] One or two sentences: what Speedboat Studio was, your role as co-founder, and the headline result (18K+ global users, $4M sales volume).
- Role
- Co-founder
- Timeline
- [placeholder] 2021–2023
- Tech Stack
- [placeholder]
- Skills
- [placeholder]
The Problem
[placeholder] What was hard for creators launching NFTs without code, and who it was for.
The Solution
[placeholder] What you built and the key decisions behind it.
[placeholder] Feature or decision
[placeholder] Short explanation.
Impact
18K+global users
$4Msales volume
[—][placeholder] metric
Finnomena - Financial Advisor Tool
A portfolio management platform for financial advisors
Finnomena is a digital wealth management platform in Thailand. When I joined, it had $1.2B in assets under advisement, 3,300+ licensed advisors, and 220,000 users. I owned the financial advisor platform, where advisors build and manage client portfolios.
- Role
- Product Manager · 2021-2022
- Industry
- Fintech / Wealth Management
- Skills
- Product Management · UX Design · User Research · Usability Testing · API Integration · Partnership
Problem
I interviewed 30 financial advisors and mapped their day-to-day workflow to find the pain points.
- Fragmented tools: One workflow took three tools, with no central place to manage a client’s portfolio.
- Manual fund selection: Advisors picked funds and allocations from their own research, with no personalized recommendations.
Solution
A centralized platform where advisors build and manage compliant client portfolios, with AI allocation recommendations from Franklin Templeton.
Centralized portfolio management
I pitched the platform to the executive team: a better advisor experience would raise satisfaction and retention, and grow AUA and fee revenue. Once approved, I led it from 0 to 1, through wireframes, usability testing, development, launch, and advisor onboarding.
- Plans in one place: Create, edit, and manage investment plans, each tied to the client’s risk profile and goals
- Automated compliance checks: Plans are rechecked when a client’s risk profile or goals change, keeping them compliant with SEC Thailand regulations
- Templates: Shared starter templates, plus custom templates advisors can save for reuse
Result: 60% adoption within two weeks of full launch. The platform is still live for all 3,300+ advisors.
AI recommendations with Franklin Templeton
Finnomena partnered with Franklin Templeton to integrate its Investment Optimization Engine through an API. I owned execution: wireframes with the design team, usability testing with 10 advisors, and the API integration with the tech lead.
- Personalized allocations: Recommendations based on the client’s risk profile and goals
- Scenario comparison: Side-by-side scenarios, such as aggressive vs balanced
Advisors still made the final call on every recommendation.
Result: Handed off the final product design to engineering for implementation.
Remark: Screens and workflows aren’t included to protect confidential information.
7-Eleven AI Commerce Chatbot
Launching the first commerce chatbot in the 7-Eleven Thailand app for 16M+ users
7-Eleven Thailand has 15,000+ stores and an app with 16M+ monthly active users. The company wanted to open a new digital sales channel as part of its digital transformation.
I was the Program Manager at Sertis, embedded with the 7-Eleven team. I managed a $450K project across business, UX, data, engineering, marketing, and customer support. This was before ChatGPT, so every intent, conversation flow, and fallback was designed by hand.
- Role
- Program Manager · 2019 to 2021
- Industry
- Online Retail
- Skills
- Program Management · Delivery · Requirements Gathering · KPI Tracking · Conversational Design · Funnel Analysis · Stakeholder Management · UAT and Go-Live
- Tools
- Google Dialogflow · SQL · Python · RegEx
Requirements gathering
The brief was to bring commerce into chat, with no detailed requirements. I analyzed app usage data to find what users did most, like browsing products, tracking orders, and finding stores. I then ran requirements workshops with the business team and defined four core flows for v1, each with clear acceptance criteria.
Scope and timeline
The business team wanted more features than engineering could deliver by the launch date. I scoped v1 to the four flows based on usage data and moved the rest to a v2 backlog. I also built buffer into the timeline for UAT fixes.
- Chat commerce: Browse a product carousel and check out without leaving the chat
- Order tracking: Check delivery order status
- Store finder: Find nearby stores, hours, and promotions
- Customer support: [Confirm the fourth flow]
Delivery
I kept six teams aligned with a shared deliverable list, clear owners, and deadlines. I worked with UX and engineering on the conversational flows, including intents, fallbacks, and scripted vs open chat. When marketing and the app team had different launch dates, I ran a joint session to agree on one. I led UAT as the single point where every team’s work came together, then managed go-live.
KPI tracking and iteration
I set up post-launch monitoring so each team could track its flow’s performance. A few weeks in, funnel analysis showed about 70% of users dropped off in the product discovery flow, which asked two to three questions before recommending a product.
I segmented session data by user intent and ran follow-up surveys. Users who already knew what they wanted were leaving. We removed the questions and let users state what they want. I aligned the 7-Eleven business owner and marketing team on the change before release.
Result: All four flows launched on time. The chatbot handled 800K+ inquiries a month with replies in under 30 seconds. After the change, drop-off fell by ~22%, and the 7-Eleven marketing executive gave us very positive feedback.
What I learned
Validate with users early: Post-launch tracking caught a problem that planning missed, and fast iteration turned it around.