Government Savings Bank · AI Laboratory · Est. 2026

GSBAIL

built in-house!

Intelligence that stays at home. Models trained, tested and served inside the bank.

We research and build language models, document and speech intelligence, and financial scoring for Thailand's national savings bank, with every byte inside infrastructure the bank controls.

See the five projects →🤗 Hugging Face

♥ x1  AI FOR EVERY THAI HOUSEHOLD ▮

0flagship projects
0Bparameters in research
0%local-first design
0customer files published

We build AI a national bank can stand behind: transparent enough to audit, robust enough for millions of customers, precise enough to earn the trust of every household.

Five projects, one stack.

all local-first ↓
01

KaoChat LLM

Local chatbot LLM

An in-house large language model assistant that serves chat on the bank's own infrastructure. Every conversation stays inside the organization's security boundary.

On-prem LLMThai languageBank infrastructure
active
02

KaoOCR

Local OCR LLM

Reads and extracts information from financial documents entirely within the organization. Built for heterogeneous layouts, Thai text and numerically sensitive fields, with confidence scoring and human review for uncertain results.

Domain-adaptiveLocal inferenceConfidence-aware
active
03

Kao AI Research Laboratory

Research LLM fine-tuning · 27B

Our core research program: investigating and fine-tuning a 27-billion-parameter language model for banking. The foundation for the lab's next generation of in-house models.

27B parametersFine-tuningOpen research
research
04

KaoLEAF Score

Financial scoring using AI

Learned Embeddings for Automated Feature representation. LEAF learns merchant financial behavior from transaction histories, keeping behavior representation, credit-outcome prediction and lending policy as separate, auditable layers.

Temporal embeddingsCredit scoringInterpretable

Project page ↗

active
05

KaoASR

Local Thai speech recognition

Domain adaptation of Thai automatic speech recognition models for privacy-constrained banking environments. Models are tuned to the language of banking while audio and transcripts stay inside the bank's own infrastructure.

Thai ASRDomain adaptationPrivacy-constrained
research

Three principles.

more than accuracy
01 · TrustTrustworthy

Rigorous evaluation, fairness auditing and governance from day one, not bolted on after deployment.

02 · ScaleScalable

Systems built for national-scale workloads, from chat to transactions.

03 · ClarityInterpretable

Outputs that can be explained to regulators, auditors and the customers behind every number.

Open models & artifacts

Find us on Hugging Face.

Checkpoints and research artifacts from Dr. Teerapong Panboonyuen are published here as the lab's work matures. Only public, authorized artifacts are released; customer data and internal documents are never published.

🤗 Open profile ↗
🤗 kaopanboonyuen
huggingface.co/kaopanboonyuen
// more soon…
Dr. Teerapong Panboonyuen
Founding head of lab

Dr. Teerapong Panboonyuen

ดร. ธีรพงศ์ ปานบุญยืน (ดร. เก้า) · Vice President of AI

Dr. Kao leads GSBAIL's strategy and research direction, setting the technical and ethical standards for AI across Thailand's national savings bank. Creator of KaoChat, KaoOCR and the KaoLEAF framework, with a focus on systems that are rigorously evaluated, transparently governed and genuinely useful at scale.

Full profile ↗🤗 Hugging Face ↗

My Day 1 at GSB.

Mon · 10 Aug 2026 ✿
$ git commit -m "day 1: hello, world"[main 0x01] welcome_party 100% ✓
The GSB team gathering in the corridor to welcome Dr. Kao on his first day
the welcome crewfig.01
Dr. Kao with his boss, holding a smiling flower balloon
my boss & me · LGTM ✓fig.02
Team members smiling in the office, some in inflatable costumes
good vibes: loaded ✓fig.03
The whole team posing around the meeting table, with a green monster costume
the whole squadfig.04
Whiteboard covered in handwritten welcome messages for Dr. Kao in Thai, Japanese, French, Korean and Italian, with a cartoon cat and the words Welcome to Data & AI Teamth 0.93th+en 0.98th 0.96th 0.94th 0.99th 0.97th 0.95th 0.93th+en 0.98th 0.96th+en 0.94ja 0.99th 0.97ja 0.95flag·jp 0.93flag·fr 0.98fr 0.96flag·kr 0.94ko 0.99flag·it 0.97it 0.95th 0.93sun 0.98sketch 0.96th 0.94th 0.99sketch 0.97th 0.95th 0.93th 0.98th 0.96en 0.94sig+date 0.99th 0.97cat 0.95th 0.93sig 0.98th 0.96lips 0.94th 0.99ball 0.9741 regions · 6 langs · μ 0.96thjafrkoiten
the welcome board
welcome.i18n.yaml
# whiteboard · Data & AI Team · 10 Aug 2026
th: "ยินดีต้อนรับ"
ja: "いらっしゃいませ"
fr: "Bonjour, bienvenue"
ko: "어서오세요"
it: "Benvenuto Dottor Kao"
notes: 20+ # all handwritten
cat:   1 🐱 # unregistered contributor
regions: 41 # auto-detected

$ pytest welcome/
> 5 locales passed ✓

Day one, and the whole team showed up to say hello. Thank you, boss and teammates ♡ I'll keep building good AI, release after release: our own local, modern AI.

AI in Banking, live.

Tue · 6 Oct 2026 ★

Speaking for GSB to a banking delegation from Sri Lanka on their study visit to the Head Office.

Dr. Teerapong Panboonyuen speaking to the Sri Lankan delegation
AI in Banking, mid-talkfig.01
talk.yaml
# study visit · GSB Head Office
topic:    "AI in Banking"
date:     2026-10-06 # Tue
guests:   "Sri Lanka delegation"
program:  "Branch Leadership"
venue:    "Ramjitti Room"
stance:   local-first ✺
status:   delivered ✓

$ echo "ayubowan 🇱🇰"
$ open gsb.or.th/news ↗
Group photo of the Sri Lankan delegation and GSB hosts on the staircase
delegation & hostsfig.02
Delegates listening in the Ramjitti meeting room
Ramjitti Roomfig.03
Wide view of the full meeting room during the session
full housefig.04

Publications, soon.

not yet published ✎

Placeholder BibTeX entries, marked "in preparation". Titles and details may change before publication.

@unpublished{panboonyuen2026kaochat,
  title  = {KaoChat: A Local Large Language Model Assistant for
            Bank-Controlled Infrastructure},
  author = {Panboonyuen, Teerapong},
  year   = {2026},
  note   = {Manuscript in preparation, not yet published}
}
@unpublished{panboonyuen2026kaoocr,
  title  = {KaoOCR: A Domain-Adaptive Local {OCR} Framework for
            Organization-Specific Document Recognition},
  author = {Panboonyuen, Teerapong},
  institution = {Government Savings Bank},
  year   = {2026},
  note   = {Manuscript in preparation, not yet published}
}
@unpublished{panboonyuen2026kaoresearch27b,
  title  = {Domain Adaptation of a 27B-Parameter Language Model
            for Banking: A Research Program},
  author = {Panboonyuen, Teerapong},
  year   = {2026},
  note   = {Manuscript in preparation, not yet published}
}
@unpublished{panboonyuen2026kaoleaf,
  title  = {KaoLEAF: Knowledge-Augmented Optimization and Learned Embeddings 
             for Automated Feature Representation},
  author = {Panboonyuen, Teerapong},
  year   = {2026},
  note   = {Manuscript in preparation, not yet published}
}
@unpublished{panboonyuen2026kaoasr,
  title  = {KaoASR: Domain Adaptation of Thai Automatic Speech Recognition
            Models for Privacy-Constrained Banking Environments},
  author = {Panboonyuen, Teerapong},
  year   = {2026},
  note   = {Manuscript in preparation, not yet published}
}