The need
Feeding Thai documents to ChatGPT or Claude wrecks table structure, and RAG over scanned Thai PDFs stays broken — the poster's free Thai OCR model exists precisely because this fails repeatedly.
Original post — verbatim Thai
ใครให้ ChatGPT หรือ Claude อ่านเอกสารไทยแล้วตารางเละ หรือทำ RAG จากไฟล์สแกน ลองตัวนี้ครับ
Three business opportunities
OPPORTUNITY 01
Hosted Thai DocParse API
- What
- A hosted API converting Thai PDFs and scans — tables, forms, receipts — into clean structured markdown/JSON ready for Claude, with accuracy SLA.
- Who pays
- SaaS teams and system integrators building Thai document workflows pay for reliability and support over free models.
- Price
- ฿0.5–2 per page metered (hypothesis)
- Why now
- Free local Thai models now exist (the post: 0.39s/page, 4MB, no GPU) — the missing layer is hosted convenience and integration.
- Effort
- M
- Moat
- A growing dataset of Thai layout edge cases and corrections improves parsing steadily.
- Risk
- Authors of free open models extend into hosting and commoditize the layer.
OPPORTUNITY 02
RAG-in-a-box for Thai scanned archives
- What
- Productized service: end-to-end ingestion plus Thai-language Claude chat over a firm's scanned archive, deployed on-prem for confidentiality.
- Who pays
- Law and accounting firms that must keep client files on their own machines.
- Price
- ฿49,000 setup + ฿5,000/mo (hypothesis)
- Why now
- Scanned paperwork remains the norm in Thai bureaucracy, and local cheap models make SMB budgets realistic in 2026.
- Effort
- L
- Moat
- Vertical parsing templates and accumulated corpus knowledge per profession.
- Risk
- Slow enterprise sales cycles strain a 3-person team.
OPPORTUNITY 03
Thai parsing benchmark + Claude skill pack
- What
- A public Thai document-parsing benchmark plus a Claude Code skill that picks the right parser per document type and reports expected accuracy.
- Who pays
- Developers pay to skip painful parsing trial-and-error.
- Price
- ฿299 one-off per skill pack (hypothesis)
- Why now
- Skills are this community's main distribution format — the post itself shares models via a group post.
- Effort
- S
- Moat
- Benchmark credibility becomes the traffic magnet and standard-setter.
- Risk
- Low price ceiling; functions mostly as a funnel, not a business.