Reasoning Engine
ExpertArch
We translate technical detections into human-level reasoning. Automated expert commentary, repair insights, and cost prediction.
The problem
The file is assembled. Somebody still has to read it.
By the time a claim reaches an assessor it holds dozens of photographs, a damage report, a policy, a registration, and a form somebody filled in by hand.
Much of that reading goes on whether the pictures and the paperwork agree. Then somebody writes it up: the risks, the repair, the likely cost, and what is still missing.
In and out
One claim, read end to end, and written up.
- A claim file
- Photographs and paperwork as they arrive — on their own, or alongside findings the other workflows have already produced.
- A question
- Whatever the reviewer wants to know, plus any photograph or document they attach.
- The file, read
- Damages, documents, risks and gaps, in the order that matters to the decision.
- Contradictions
- Where the photographs, the findings and the documents disagree, listed as points to verify.
- The write-up
- Verified, uncertain and follow-ups required under their own headings, with the repair route and the likely cost.
Every line points back to the evidence behind it: the photograph, the finding or the field it came from.
Capabilities
Multi-modal reasoning
Synthesize insights from documents, images, and metadata into a coherent assessment.
Damage-to-repair mapping logic
Apply logic to determine optimal repair actions and labor estimation.
Risk modeling & Summary generation
Model mechanical risk probability and generate natural-language expert summaries.
The hard part
Pointing an assistant at raw photographs is the weak version.
Mixed evidence, one reading
Photographs, findings and paperwork rarely agree on first pass. The work is holding them together and saying plainly which parts are established and which are still open.
Seeing it, proving it
A damage visible in a photograph and a damage backed by pixel evidence are recorded as different things, and where the two disagree the file says so.
The join nobody wants
Identity, capture quality, damage and documents arrive together with their confidences intact. Kept in four separate answers, the comparison falls back to a person doing it by hand.
Where a person decides
It writes the assessment. The judgement stays with you.
Uncertainty is part of the answer
Where the evidence does not carry a conclusion, it says so, and it names the photograph or field that would settle it.
Thresholds are yours to set
Every reading carries its confidence, so straightforward files can move on their own and the rest reach a person with the reason attached.
It says when it cannot
When a plate cannot be read or a document cannot be graded, the file records that rather than a plausible guess.
Who it is for
Insurance
Motor claims teams that want the file read, the risks named and a first assessment written before assignment.
Fleet and rental
Handover and return files checked for the evidence a dispute will need, and what the damage will cost.
Inspection and remarketing
A second reading of every condition file, so the report leaves with its gaps already named.
How it fits
Start with one file, then shape the process around it.
Open the file
The assessment is already there. The conversation starts against that claim, with its photographs and findings loaded.
Point and ask
Select a damage finding or a document field on the report and ask about that one thing.
Set it up once, in the console
Define what a file should yield and how it moves through your process, then let it run on every claim after that — without writing code, and versioned so a change is reviewable.
The assessment and the exchange behind it stay with the claim, searchable, so a later review can see how it was reached.
Questions we get asked
What is ExpertArch?
ExpertArch is Archmir's reasoning engine. It reads one assessed claim: the photographs with the damage marked, the vehicle details, the findings and the extracted document fields. It writes the assessment: what is verified, what is uncertain, the risks, the repair route and the likely cost.
What do I need to send?
A claim that has already been through the assessment workflows. From there you ask questions, and you can attach further photographs, a document or a screenshot you have marked up.
Does it replace the assessor?
No. It removes the reading pass and the cross-check, so the assessor opens a written assessment rather than a folder. Where files clear automatically, you set that threshold.
How do I know it is right?
It is written from findings that carry their own evidence and confidence, and it separates what the file proves from what it only suggests. We publish no accuracy figure; test it on your own closed files.
What does it not do?
It stays inside one claim: questions outside the file in front of it are refused, and it prepares the assessment rather than issuing the decision. Where files clear automatically, that threshold is yours to set.
Where to start
The assessment you would have written by hand is the one to compare it against; start with a single claim.
Initialize ExpertArch