The private AI layer for property risk.

A grounded, citation-or-refuse AI portal for CriticalAsset, InsureMEP, and the carriers, brokers, and inspection partners we work with. Connect your own documents and customer records, ask questions in plain English, and get answers backed by the source material — never hallucinated.

Open the portal How it works

Sign in with Google. Open to staff and approved partners. Your data stays in our Google Cloud project, encrypted with our keys, and is never sent to public AI tools.

Status: Live

Region: us-west1

Model: gemini-2.5-pro

What it does

The portal answers questions over your documents and records. Instead of guessing, it retrieves the chunks from your sources that actually match the question, then writes the answer using only those chunks — and shows you the citations. If nothing relevant is retrieved, it returns UNSUPPORTED rather than inventing an answer.

Streaming chat with citations: Conversational Gemini grounded on the corpora you select. Every factual claim links back to a chunk you can click through to the source. Multi-turn context within a session.
Report QA Copilot: Verifies every factual claim in a Water Containment draft against the cited source evidence. Returns CITED, UNSUPPORTED, or AMBIGUOUS per claim. The same citation-or-refuse contract that powers chat, applied to report-level QA before you ship.
Google Drive ingestion: Drop documents into a "CriticalAsset Ingestion" folder we create in your Drive. The portal pulls them in (PDFs, Docs, Sheets, Slides, plain text) and indexes them into the corpus you choose. Subfolders are walked recursively; updates re-ingest automatically.
Capture (voice + text): Dictate or type what happened on a customer call. The portal structures the note and syncs it to HubSpot. Saved locally first; sync runs in the background.
Direct upload + admin: Upload PDFs, DOCX, MD, HTML directly into a corpus. Manage ingestion state, retry failed jobs, soft-delete documents. Account corpora sync from HubSpot when that connection is on.
Pricing engine: Rule-driven indication calculator for property submissions. Runs on a versioned, citable rule plan; every indication pins a corpus snapshot so you can re-run any prior quote bit-for-bit.

Who it's for

This portal is built for the people who actually work property risk across the CriticalAsset and InsureMEP teams and the partners that extend that work:

CriticalAsset and InsureMEP staff — Water Containment analysts, Product, Customer Success, Pricing, and the engineering team keeping the lights on.
Carrier and broker partners — when invited, partners sign in with their Google Workspace account and get scoped access to the corpora and tools the relationship covers.
Inspection and underwriting partners — companies that generate the source material the portal grounds on (inspection reports, building schedules, claims history).

Access is auditable. Every sign-in, query, retrieval, and document upload is logged with the user's email and a trace ID. Per-corpus membership controls who sees what.

Why it exists

Knowledge was scattered across public AI tools. Water Containment analysts used one chatbot. Product used another. Pricing guides lived in Drive. Customer interactions lived in HubSpot. Every team was individually leaking proprietary context into models outside the company perimeter, and no answer anyone got was grounded in our own data.

This portal is the consolidation. One private place to ask questions, backed by sources you control, with the audit trail and access controls a serious property-risk operation needs.

Privacy & data handling, in brief

Full details on the Privacy Policy page; the short version:

Documents you connect (Drive, HubSpot) and upload (Admin → Content) are stored in our Google Cloud project, encrypted at rest with our own KMS keys (CMEK, 90-day rotation).
We never send your data to a public AI tool. Generation runs on Gemini 2.5 Pro inside the Google Cloud project, governed by Vertex AI's data-use terms.
Google Drive integration uses two narrow scopes: drive.file (so we can create the inbox folder) and drive.readonly (so we can read files you put in that folder). Our code only ever lists files inside the folder you connected — it does not crawl the rest of your Drive.
Every prompt, retrieval, generation, and document upload is logged to Cloud Logging with your email + a trace ID, retained for 7 years.
Disconnecting wipes the credentials. Files already ingested into your corpus stay there until you delete them via Admin → Content.

Get in touch

Bug, feature request, or partner onboarding question: casey@criticalasset.com. Security concerns: security@criticalasset.com.