Skip to content

How we use AI

AI drafts. Your team decides.

A model reads your plans, code checks each quote against its page, and nothing reaches your record until a person accepts it.Your data is not used for training, and you can bring your own OpenAI key.

  1. A model reads

    • Reads scanned pages, tables and figures
    • Quotes each hazard mention word for word
    • Drafts justifications from verified evidence
  2. Code checks

    • Each quote against the page it cites
    • Each drafted sentence for a source
    • Hazard names against the taxonomy
  3. Your team decides

    • Which hazards are on your list
    • Which draft to accept, edit or rewrite
    • Each 1–5 score, and who approves it

1 · Reading the plan

First, the PDF becomes something we can check.

Each plan goes through seven steps before any AI writes. Five steps are ordinary software. They give the same result each time. AI helps with two: reading scanned pages and making the plan searchable.

  1. PDF
    a1c3…9f
    412 pages
    Opens fine
    1Software

    Check the file

    We make sure the PDF opens, count its pages and give it a fingerprint, so the same plan is never stored twice.

  2. TextScanTableMap
    2Software

    Sort the pages

    Each page is sorted: ordinary text, a scan, a table, or a map or picture.

  3. 1
    2
    3
    4
    5
    3Software

    Copy out the words

    The words and tables are copied out in reading order. Scanned pages are read with text recognition.

  4. Scans, tables, maps
    Read by AI
    4AI model

    Read the hard pages

    An AI vision model looks only at scans, tables and maps, the pages plain text can't capture.

  5. 1Introduction
    2Hazard profiles
    2.1Wildfire
    2.2Flooding
    3Mitigation
    5Software

    Rebuild the outline

    The table of contents, chapters and sections are put back together.

  6. WordsTables & mapsOutline

    One trusted copy

    Every citation points here

    6Software

    Make one trusted copy

    Everything becomes one digital copy of the plan. Quotes and footnotes point back to it.

  7. evacuation notices
    4.2 Wildfire history
    6.1 EvacuationNotices reached 61% in the first hour
    7AI model

    Make it searchable

    Each section is indexed with a short summary, so the right passage turns up in seconds.

2 · The quote check

If the quote isn't on the page, it's gone.

Code compares each quote to the passage it cites, word for word. It ignores only capitalization and spacing. A quote that is not there is discarded. The reason is logged. The model cannot override this check.

Pine Hollow Fire After-Action Reportp. 4 · p004-b002

Between 2010 and 2024 the District responded to 64 wildland fires, the majority of which burned less than one acre. The Pine Hollow Fire of August 2018 burned 1,240 acres and destroyed 14 homes.

The model says the page says

“the District responded to 64 wildland fires”

Found in block p004-b002 · kept, p. 4

The model says the page says

“the District responded to more than 80 wildland fires”

Not on the page · discarded, reason logged

3 · The writing

Writing starts only after the evidence passes.

The model receives only quotes that passed the check, facts your team kept, and federal data. A fact you exclude is not sent. A sentence that cites nothing, or cites something the model was not given, is dropped and counted.

Only what already passed

  • Verified quote

    “64 wildland fires from 2010 through 2024…” · C, p. 4

  • Verified quote

    “18,400 acres of wildland–urban interface…” · B, p. 12

  • Federal record

    FEMA NRI wildfire risk: relatively high · NRI 1.20

The model drafts

Every sentence must name what it cites

Citation check · code

The county recorded 64 wildland fires from 2010 through 2024, most under one acre.1 About 18,400 acres of interface lie along the western foothills.2

Wildfire activity in the region is expected to double by 2040.

Cites nothing it was given, so it is dropped and counted.

A person accepts, edits or rewrites it. Until then it is a draft.

What the checks can’t do. A check proves a quote is on the page. It does not prove a summary is fair. Each sentence links to a page you can open. Counter-evidence is labelled, not counted. A person must accept model text before it reaches your record.

4 · AI and data security

Your data, your controls, and a record of each step

  • GovRAMP control framework: In place
  • GovRAMP review: Not started yet

Hazzard runs on a GovRAMP control framework today. We have not entered the GovRAMP review yet, because we are in beta pilots with a small group of local governments. The controls and the audit trail below are available to each pilot now.

See subprocessors and the GovRAMP path →

hazzard.ai/o/cedar-ridge-county/settings/ai

AI account

Hazzard managedYour OpenAI key
sk-••••••••3f9aTested

Monthly budget

$310 of $500 this month · admins hear at 80%

Who can act

  • Named accounts. Each person signs in as themselves. Hazzard operators use multi-factor authentication.
  • Roles and approvals. Roles set who can read, edit, score and approve. You set how many approvals a score needs, from none to three.
  • People decide. A hazard stays a suggestion until a person includes it. Only a person sets a score.

Where your data goes

  • No training. Model calls go to OpenAI's API, which does not train on API data. Each call asks OpenAI not to keep the response, except deep research, which OpenAI keeps for up to 30 days so an interrupted run can resume.
  • Published vendors. Each subprocessor that handles customer data is listed on our security page, with its purpose.
  • Your own OpenAI key. Optional. An admin adds it, Hazzard tests it before storing it, and calls bill your account. If OpenAI refuses it, runs pause. Hazzard does not switch to its own key.

What is recorded

  • A history you cannot edit. Uploads, decisions, approvals and re-runs. Each model call is one row. Each run records its model (gpt-5.6-terra today) and prompt version (1.3.0).
  • Each claim to its page. The source page with the passage highlighted, and what the run threw away and why.
  • Spend you can cap. Settings show the month's AI use. Admins hear at 80% of the budget, and new runs wait at the cap.
Book a walkthrough

Bring a plan. Watch Hazzard read it.

Tell us about your jurisdiction. We set up the workspace, you upload a plan, and each claim Hazzard finds is cited to a page you can open.

Book a walkthrough

Takes about two minutes. A person reads each request.

Or start by marking your area on the map