Preliminary seismic screening with the source behind every value
Seismetra AI is a working environment for screening existing W1A, C1 and C2 buildings. AI prepares and cross-checks building data. A rules engine calculates the Level 1 score. A licensed engineer reviews and approves every result.
- Document
- Assessor record, p. 1: “NO. OF STORIES: 2”
- Photo
- P-02, street elevation: parking at grade
- AI note
- Record may count residential stories only
- Decision
- M. Ortiz, Engineer · justification recorded
- Score impact
- None for W1A with soil Type D (1–3 stories)
Screen one building from documents to issued report
Follow the six steps, or click through any screen. Use “Do this step” to move quickly during a presentation. Nothing you do here is saved.
From building files to an approved, versioned result
Documents for existing buildings span decades and often disagree. The platform keeps every value tied to where it came from, so the engineer can tell facts from assumptions, suggestions and gaps.
- 1Collect materials
Permits, certificates of occupancy, assessor records, drawings and site photos are uploaded to one building record with a permanent ID.
- 2AI extracts and cross-checks
Each suggested value links to a file, page and text excerpt. Photos produce observations, not conclusions. Conflicts are flagged.
- 3Engineer resolves
The engineer accepts, corrects or rejects each value, records a justification, or requests more information. Unknown soil is marked DNK.
- 4Rules engine scores
A version-controlled FEMA P-154 table gives the basic score, modifiers and minimum score. The language model explains the result but never sets coefficients.
- 5Approver signs off a version
An authorized approver reviews sources and changes, then approves or returns that specific version. Approval rights are assigned, never self-selected.
- 6Report and change control
AI drafts the report and the engineer approves the text. New data creates a new version for review. Issued reports stay in history.
What the software does, and what stays with the engineer
Suggested values never enter the verified record without an engineer's action. Accepting one field does not approve the screening.
Calculations run in a separate rules library with test cases. A change of prompt or model cannot change a number when inputs and methodology are the same.
File, page, excerpt, original notation and converted units are stored with each value, along with who decided and why.
A material input change creates a new version and sends the dependent result back for review. Earlier approvals and issued reports remain on record.
- a percentage of building safety or an earthquake prediction;
- a determination of whether a building is fit for occupancy;
- a decision on whether a municipal retrofit program applies to an address;
- a substitute for the detailed evaluation that a program or engineer requires.
Built for engineering firms first
Screen portfolios of W1A and C1/C2 buildings, share review between engineers and approvers, and issue consistent reports.
Receive approved results and a clear order for further evaluation through the engineering firm that serves them.
Building inventories and priorities for further assessment, with methodology, access and reporting agreed in advance.
Run a pilot on your own W1A or C1/C2 buildings
Before the pilot starts, we agree on the buildings, the available materials, the responsible engineers and the acceptance criteria. The pilot then measures:
- time to prepare building records and reports;
- share of AI suggestions the engineer had to correct;
- completeness of source references;
- number of comments raised during review.
We make no claims about accuracy or time savings until pilot results are available.
Los Angeles, California