About Archmir
We build the reasoning layer between raw visual input and an operational decision.
What we build
Most computer vision products stop at detection. They tell you a bounding box exists and leave the interpretation — is this repairable, what does it cost, does the paperwork match the vehicle — to a human reading a dashboard.
Archmir is built around the opposite premise: detection is the cheap part. The expensive part is the reasoning that turns detections into a decision an operations team can act on without re-doing the work.
How the platform is arranged
The platform is five workflows over one pipeline. Each is independently callable, and each produces structured output the next stage can consume.
| Workflow | Input | What it resolves |
|---|---|---|
| CarScope | Vehicle photographs | Plate, VIN, windshield markings, make/model/trim/colour, and pixel-level masks for 30+ exterior parts |
| ImageScope | Any photograph | Scene and weather context, lighting, blur and reflectivity, camera angle, proximity and orientation |
| DamageScope | Vehicle photographs | Damage segmentation, type and severity, repair-versus-replace decisions and labour estimates |
| ArchDoc | Documents | Classification across 30+ document types plus AI-first field extraction, beyond template OCR |
| ExpertArch | All of the above | Multi-modal reasoning, risk modelling and natural-language expert summaries |
The separation matters operationally. A claims workflow that only needs vehicle identification does not pay for damage segmentation, and a document pipeline does not carry a vision model it never calls.
Engineering principles
Strict contracts over convenience. Every response is JSON with a stable, versioned shape and per-field confidence scores. A model improvement must never require a client change.
Confidence is part of the answer. A prediction without a calibrated confidence score cannot be routed automatically. Every extracted field carries one, so downstream systems can decide what to auto-approve and what to escalate.
Idempotency by default. Batch jobs accept idempotency keys and report through webhooks. A retried request after a network failure produces the same job, not a second one.
Infrastructure
Archmir was selected for Google for Startups' Scale AI tier, which provides dedicated cloud resources, premium technical support and enterprise-grade AI infrastructure. The platform runs on that foundation.
Compliance
Archmir Teknoloji A.S. processes personal data under Turkish Law No. 6698 (KVKK) and the EU General Data Protection Regulation. Processing purposes, lawful bases, retention periods and data subject rights are documented in full in the Privacy and Information Security Policy, and the commercial terms in the Corporate Terms of Service.
