AXCXEPT
Contact Us
Services Strengths Process Case Studies About Contact
Domain Data flows into a fine-tuned LLM deployed in an air-gapped environment

AXCXEPT Inc. — Sapporo, Japan

Domain-Specific LLM Fine-Tuning
& Air-Gapped AI Deployment

Custom LLMs trained on your industry data. Deployed in closed environments where data never leaves your premises.
Our trusted partner teams handle everything — from PoC to production.

25+
LLMs Published (OSS)
¥50M+
Enterprise Contracts
22yr
Enterprise IT Experience
CKA/CKAD
Kubernetes Certified
Request a Consultation View Services ↓

Services

End-to-end AI solutions delivered by our trusted partner teams.

LLM Fine-Tuning

Domain-Specific LLM Fine-Tuning

We fine-tune open-weight LLMs on your proprietary domain data — legal, medical, manufacturing, finance, and more. Our partner teams handle data curation, training pipeline setup, evaluation, and model delivery.

  • Full-parameter / LoRA / QLoRA fine-tuning
  • Domain-specific evaluation benchmarks
  • EZO series heritage (25+ published models)
Air-Gapped AI Environment

Air-Gapped AI Environment

Deploy LLMs in fully closed networks — no data leaves your premises. We design and build on-premise or private cloud AI infrastructure with Kubernetes orchestration and GPU optimization.

  • Zero external data exposure architecture
  • On-premise GPU / private cloud deployment
  • Kubernetes-native (CKA/CKAD certified)
AI Consulting & PoC

AI Consulting & PoC

Not sure where to start? We assess your AI readiness, design proof-of-concept projects, and validate ROI before committing to full-scale deployment. Strategy grounded in hands-on engineering.

  • AI readiness assessment
  • PoC design, execution & evaluation
  • Production roadmap & architecture design

Why AXCXEPT

Deep LLM expertise meets enterprise delivery discipline.

Proven LLM Engineering

25+ open-source LLMs published on HuggingFace. Our EZO series achieved GPT-4o-class performance on Japanese benchmarks. This isn't theory — we build and ship production models.

Trusted Partner Teams

We assemble dedicated project teams from our trusted partner network. Each engagement gets a purpose-built team with the right skills — infrastructure, ML engineering, domain expertise — so delivery is seamless.

Air-Gap Specialists

Security-critical industries demand closed environments. We design zero-data-leakage architectures with on-premise GPU clusters, private registries, and Kubernetes orchestration — CKA/CKAD certified.

PoC to Production

We don't stop at consulting decks. From initial PoC through evaluation, optimization, and production deployment — our teams own the full lifecycle. Over ¥50M in enterprise contracts delivered.

Delivery Process

From first conversation to production deployment.

1 2 3 4

Consultation

Understand your business goals, data landscape, and security requirements. Define scope and success criteria.

PoC & Validation

Build a proof-of-concept with your data. Evaluate with domain-specific benchmarks. Prove ROI before scaling.

Team Assembly

Assemble a dedicated team from our trusted partner network. Right-size for scope — ML engineers, infra specialists, domain experts.

Deployment & Support

Deploy to production — cloud, on-premise, or air-gapped. Ongoing monitoring, model updates, and operational support.

Track Record

Selected achievements from our consulting and LLM engineering work.

Enterprise RAG System

Delivered a large-scale RAG implementation for a major Japanese corporation. 2,819 pages processed with +15% precision improvement and +10.8% faithfulness gain. Over ¥50M in contracts within 6 months.

EZO LLM Series

Developed 25+ Japanese LLMs published open-source. EZO-70B achieved top scores on ElyzaTasks-100. EZO-8B scored MT-Bench 9.08 — GPT-4o-class performance on consumer 24GB VRAM hardware.

Global Recognition

Featured in Alibaba Cloud Official Customer Case Study. Speaker at JUMPSTARTER 2025 (Hong Kong) and Alibaba Cloud Tech Day Tokyo. Published research on low-cost LLM training methods.

News

2025.04.15

Released "Pseudo GRPO/PPO approach for low-cost Japanese LLM performance improvement"

2025.04.10

Released ultra-compact LLM "QwQ-32B-Distill-Qwen-1.5B-Alpha" specialized for math reasoning

2025.03.26

Panel discussion at ASIAWORLD-EXPO, Hong Kong (Jumpstarter)

2024.12.10

Speaker at ALIBABA CLOUD TECH DAY TOKYO

About

CompanyAXCXEPT Inc.
CEOKazuya Hodachi
BusinessDomain-Specific LLM Fine-Tuning, Air-Gapped AI Environment Deployment, AI Consulting & PoC, LLM Development (EZO Series)
LocationHondori 14, Shiroishi-ku, Sapporo, Hokkaido, Japan
DeliveryTrusted partner team-based project execution

Founder

KH

Kazuya Hodachi

CEO & Founder, AXCXEPT Inc.

22 years in enterprise IT — from infrastructure to AI. Built teams, shipped products, and led multi-million dollar projects at leading global technology companies. Now leads AXCXEPT's vision, assembling trusted partner teams to deliver domain-specific LLM solutions for enterprises.

Credentials

Azure Solutions Architect Expert CKA / CKAD 25+ LLMs on HuggingFace

Recognition

Speaker, JUMPSTARTER 2025 (AsiaWorld-Expo, Hong Kong)
Speaker, Alibaba Cloud Tech Day Tokyo (2024)

"We don't just advise — we assemble the right team, build the solution, and deliver measurable results."

Contact

Let's discuss how domain-specific LLMs can transform your business.