Product division — Institute of LAG Research
AI adoption is often framed as a trade-off between efficiency and safety: move fast and you court risk; move carefully and you stall.
But you don't have to choose. SVELGR introduces AI through accumulated scientific knowledge and refined UX design, then keeps questioning it, rigorously. Integration, development, and audit — all within one research institute.
01Services
Putting AI into a business and confirming that the AI is safe are not separate jobs. SVELGR treats them as one.
ABuild
We support the integration of AI into operations and products end to end — from requirements through design, implementation, and operation. Safety is not added afterwards; it is where the design starts.
For example
BVerify
We examine AI systems — in development or already in production — from an independent standpoint. We surface concerns, rate their severity, and set out what to address first.
For example
These are examples. Scope and approach are designed around what you bring to the conversation.
02Audit scope
AI risk does not sit in a single domain. We examine the whole system, across technology, data, and organization.
Prompt injection, data leakage, privilege escalation: we test the attack surface that is specific to AI.
How personal and confidential data is handled across training, inference, and logging.
We measure hallucination and output variance, and question whether the evaluation design itself is sound.
Whether outputs are skewed, and whether the basis for a decision can be explained.
Alignment with internal policy, and with guidelines and regulation in Japan and abroad.
Whether monitoring, incident response, and human oversight actually work in production.
03Approach
Integration and audit follow the same discipline. At every stage we make clear what is known and what remains open.
01
Map the current operations, systems, and data; clarify goals and constraints.
02
Define the architecture and the evaluation criteria, safety requirements included.
03
Build small, verify repeatedly, and raise it to production quality.
04
Test and audit across security, quality, and governance.
05
Keep the system safe through monitoring and continuous improvement.
04Products
What does it actually take to integrate AI safely? SVELGR also puts its answer into practice in a product of its own.
P-01Memori
Coming soon to the App Store
Baby-care tracking app — iPhone, iPad, Apple Watch
An app for logging feeds, sleep, temperature, and growth, for keeping photos and video, and for sharing them with family. From the accumulating record, AI finds patterns and predicts the next feed or nap. Its design carries the same decisions SVELGR recommends to businesses.
Visit memori.babyDesign decisions
Prediction, pattern detection, and person recognition in photos run entirely on-device, with Core ML.
Data supports research only on the plan where users opt in to that, and only as anonymized statistics. The plan without research participation encrypts posts and albums end to end.
No ads, and personal data is neither sold nor shared. The service is funded by monthly plans.
05About
SVELGR is the product division of Institute of LAG Research. LAG is a research institute working on computational medicine, neuro-symbolic AI, and related fields; SVELGR brings the knowledge and the habit of verification developed there into its own products, and into the places where businesses actually use AI.
Principles
Not bolted on once the features work — built into the first design decisions.
Instead of saying a system is safe, we show it with test and evaluation results.
Methods developed in research, applied to quality assurance in the field.
06Contact
Whether you are considering AI adoption or uneasy about a system already in production — conversations at the stage where the problem is not yet well defined are welcome.