Research & Insights
Our Technology
Model Merging
Model Merging
Combine the strengths of several specialist models into one new model.
- Merge specialist models without retraining from scratch
- Merge recipes tuned automatically for each goal
- Keep the skills of every source model
Small Language Models
Small Language Models
Compact models trained deeply on one domain.
- Lower cost to train and to serve
- Easier to audit, update and replace
- Fit for private and regulated environments
Multi-Model Orchestration
Multi-Model Orchestration
A coordination layer that sends each task to the best-fit model.
- Route requests by task, language and cost
- Combine answers from several models
- Fall back gracefully when a model is unsure
Japanese-Language AI
Japanese-Language AI
Models that treat Japanese as a first language, not a translation.
- Native handling of Japanese text and business writing
- Respectful language (keigo) and context awareness
- Bilingual workflows across Japanese, English and Vietnamese
On-Device & Edge AI
On-Device & Edge AI
Models small enough to run where the data lives.
- Runs on local servers and edge hardware
- Sensitive data stays on premises
- Low latency without a round trip to the cloud
Safety & Evaluation
Safety & Evaluation
Every component, and the system as a whole, is measured.
- Evaluate each model and the combined system
- Traceable lineage for every merged model
- Human review built into the release process
Where We Work
- JapanPrimary market
- VietnamEngineering & partnerships
- GlobalOpen research & collaboration