Main site

Construction AI agent · Stormwater

Construction Stormwater Compliance AI Agent

Track the rain trigger your own permit uses, test each inspection report against the elements that permit enumerates, and hold the file for the credentialed inspector and the signatory who certifies it.

Typical delivery
4–6 weeks
Deployment
Your stack
Two signatures
Pre-submission
After launch
Agent Care

What this agent does

Assembles the record, never certifies it

In

  1. 01

    A site is enrolled from supported permit, portal or project sources, with the coverage it actually holds.

  2. 02

    A permitting authority is named per site, and its own trigger, element list and clocks are carried forward.

Reason

  1. 03

    Rain falls or a forecast crosses that permit threshold, and the inspection it has just made due is raised.

  2. 04

    An inspection falls due, and the credential of the person named to perform it is checked against that date.

  3. 05

    A report comes back, and it is tested element by element against the list its own permit enumerates.

Decide

  1. 06

    A condition is found, and the corrective-action clock is counted from discovery, not from the next report.

  2. 07

    A gap stays open, and the agent names the missing element instead of completing the record around it.

Out

  1. 08

    A site closes out, and the file is retained against the permit expiry date rather than practical completion.

  2. 09

    Execute write actions only inside the approval boundaries agreed during implementation.

Product statement

The agent assembles the record; the credentialed inspector signs the inspection, and the 40 CFR 122.22 signatory certifies the submission.

Example workflow

One site, rain to retention

  1. 1Site enrolledPermit coverage, permitting authority, site map and stormwater team
  2. 2Obligations assembledInspection trigger, credential register, corrective-action clocks and report elements
  3. 3Inspection raisedTrigger, due window, named inspector and credential status
  4. 4Controls appliedElement-by-element report checks, credential-expiry checks, clock checks and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is filed at any of them — the agent is assembling, and the signatory lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the permit signatory to certify.

Low confidence

Adds a credentialed read first.

Signatory certification

The file is held with its report elements, its open clocks and the confidence.

Accept · Amend · Return to inspector

Certified — released to file

  1. 6Permit records updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedMissing elements, clock breaches, credential lapses and post-filing corrections

Amendments

Every signatory amendment is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Performing the site inspection the permit requires.
    • Certifying any submission under 40 CFR 122.22.
    • Determining that a discharge complies with the permit.
    • Attributing an observed sheen or deposit to this site.

  • Automation boundary

    Agent acts unaided

    • ✓Track the trigger, the forecast and the inspection now due.
    • ✓Check each report against the elements its permit enumerates.
    • ✓Count the corrective-action clock from the time of the discovery.
    • ✓Name what is missing, and hold the file for the two signers.

    Any write happens inside the boundaries agreed at implementation, never ahead of certification.

    • Signing the asbestos NESHAP notification certification.
    • Deciding that a corrective action is complete.
    • Filing the Notice of Intent or Notice of Termination.
  • Changes to permit, trigger or signatory authorisation.

Example output

One inspection report, annotated

Everything the agent assembles is attached to the record it was drawn from.

Report-element output · single inspection

Illustrative example
Site
Grading, phase two
Assembled line
Post-precipitation inspection raised on the gauge reading
Recorded rainfall
0.62 in
Trigger source
On-site rain gauge
Confidence
91%
Signatory
Named on file with the authority

As received

Taken from the permit on file and the site rain gauge — nothing on this side is written by the agent.

Source records used

  • Permit element list
  • Rain gauge log
  • Credential register

Why this is held

The remedial-action schedule is blank — an element the permit lists and this report omits.

What the score decides

Below the configured threshold the file picks up a credentialed read before it.

Action

AcceptAmendReturn to inspector

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

Every site

From the permit on file

03Assembly

Assemble from the permit

Draw on the coverage that site actually holds — federal, state or tribal — with its own trigger and element list.

01Approved path

Rain starts the clock

Weekly and weather-triggered inspections are raised before the window closes.

02Human review

Send the read to what is missing

Absent elements, lapsed credentials and running clocks are marked, so the read starts where the exposure sits.

04

Build an evidence trail

The inspection, the condition it found and the signatory who certified stay on the file.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Permit and filing systems

    • EPA NeT
    • NPDES eReporting
    • SMARTS
    • state portals

  • Weather and rainfall

    • NWS forecast feeds
    • On-site gauges
    • telemetry

  • Site and field capture

    • Procore
    • Autodesk Build
    • Field inspection apps

Agent

Stormwater compliance

  • Reads the permit
  • Raises the inspection
  • Holds for signature

  • Credentials and training

    • CASQA QSD / QSP records
    • Ecology CESCL records

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six layers between the model and the permit file

Six sieves, coarsest first. What falls through all six is set out in the map beneath this.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeRevert to report assembly when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, permit-library and trigger-configuration changes.Track
L4TraceabilityRecord the permit read, the elements tested, the clocks and the signature time.Record
L3Signatory sign-offHold files for the named signatory; it governs filing, not whether the report is right.Gate
L2Policy guardrailsTest files against the element list in the permit in force and the credential register; a failure returns the file.Restrict
L1Confidence thresholdsRoute low-confidence files to a credentialed read before the signatory sees them.Require review
Model coreFile assembled — inspection record, elements tested, open clocks and confidence
L1 – L2
Test whether a file may stand
L3
Puts the filing in the hands of a signatory
L4 – L5
Keep the inspection and the condition behind it
L6
Reverts to report assembly when signals degrade

How Nestack evaluates it

Evaluate the compliance workflow — not only the finished report.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the record the authority reads

  1. E1Final-output evaluationDid the assembled report carry each element that permit enumerates?
  2. E2Step-level evaluationDid the agent use the right permit, trigger and credential register?
  3. E3Tool evaluationDid it read and write the correct site and the correct inspection?
  4. E4Confidence calibrationDo low-confidence files actually attract more signatory amendments?
  5. E5Slice evaluationHow does performance change across specific site and condition types?
  6. E6Business outcomeHow many files needed an amendment or a correction after filing?

Floor — the outcome the permit holder answers for

Failure modes

Where each failure originates in the agent

Seven failure modes, each set at the lifecycle stage where it originates.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • FZ-03

      Wrong permit template

      Site read against a permit it does not hold.

    Stage gathers

    Permit coverage, triggers, credentials and forecast

  2. 02 · Reasoning

    2 modes

    • FZ-04

      Trigger misread

      A forecast-based trigger is read as a measured one.

    • FZ-06

      Causation asserted

      A drafted line attributes an observed condition to the site.

    Stage proposes

    Inspection record, elements tested and confidence

  3. 03 · Tool / write

    2 modes

    • FZ-02

      Lapsed credential accepted

      Inspection recorded against an expired certificate.

    • FZ-05

      Clock started late

      Corrective-action clock run from the report, not discovery.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • FZ-01

      Required element missing

      Report omits an element the permit enumerates.

    Stage returns

    The record the signatory certifies and

  5. 05 · Change / Version

    1 mode

    • FZ-07

      Silent permit regression

      A permit or rule change widens what the agent will assert.

    Stage tracks

    Model, prompt, permit library and trigger config

  • Sev-1 · filed outside the boundary
  • Sev-2 · an incomplete report is certified
  • Sev-3 · source degrades, file routes to review
See the complete Construction failure-mode catalogue

Affected slices

Site-level quality can hide one bad site type

A site-level report-element completeness figure can read well while a few site types absorb most of the missing elements. Nestack reports the incomplete-report rate by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Forecast-triggered sites8.8%3.7×Review
Multi-operator sites6.3%2.6×Review
Post-event conditions4.5%1.9×Watch
Weekly routine inspections2.1%0.9×Normal

Bar: incomplete-report-rate lift vs. weekly-routine baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

Nothing closes without a new case

The loop shuts when the missed inspection is a regression case. That suite is what the next report filed is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Incomplete-report rate rises on one site type.

  2. 02

    Diagnose

    The storm that fell on a Saturday is read back through the file until one cause is left.

  3. 03

    Improve

    Any change goes out numbered, with the sites that caused it attached.

  4. 04

    Verify

    One inspection case still failing is enough to hold the release back.

  5. 05

    Learn

    One case joins the suite, one line joins the site record.

  6. Learn → Detect

    The return edge. The next storm is met by a suite one case longer.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, permits, inspection workflow, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Permit workflow discovery and boundary setting.Week 1
  2. 02Weather and credential source assessment.Week 1, Week 2
  3. 03Permit-library, trigger and element-list mapping.Week 1, Week 2
  4. 04Site enrolment and coverage normalisation.Week 2
  5. 05Trigger logic and report-element binding.Week 2, Week 3
  6. 06Confidence scoring and clock routing.Week 3
  7. 07Signatory certification workflow.Week 3, Week 4
  8. 08Permit-portal and weather integration.Week 5
  9. 09Trigger and corrective-action cases.Week 4, Week 5
  10. 10Guardrails and certification controls.Week 4
  11. 11Inspection-trail instrumentation.Week 5, Week 6
  12. 12Deployment, documentation and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne site, one permitProductionProduction permit systemsAdvancedMultiple states / permits
Introduced at Pilot
Assembly to your permit and rules✓✓✓
Signatory hold✓✓✓
Report-element baseline✓✓✓
Introduced at Production
Reporting by site—✓✓
Sign-off workflow in your systems—✓✓
Approved write-back—✓✓
Weather-and-site integration—✓✓
Introduced at Advanced
Multi-authority permit rules——✓
Multi-stage signatory chains——✓
High site count——✓
Multi-permit inspection controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, site count, approval controls and deployment requirements.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01Your permit coverages and active site list

    Site enrolment and permit-library mapping

    Week 1

  2. 02Representative inspection reports

    Assembly baseline, element extraction and record binding

    Week 2

  3. 03Your permit conditions and element lists

    Permit-library, trigger and element-list mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Weather, portal and credential assessment, then integration setup

    Week 2

  5. 05Reports you would not want subpoenaed

    Rain cases and the evaluation suite

    Week 4

  6. 06What no inspection report may leave blank

    Confidence scoring, clock routing, guardrails and certification controls

    Week 3

  7. 07Named credentialed inspectors and signatories

    Certification workflow, then pilot and production validation

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

The widths come off worked weeks rather than off the slide, which is why week five holds a pair.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Permit workflow discovery, element mapping and the automation boundary

W2Source integration and the assembly baseline

W3Inspection workflow, trigger logic and certification controls

W4Evaluation suite, element checks and failure-mode testing

W5Portal integration, pilot sites and targeted corrections

W6One rain season run under the permit signatory, then Agent Care handover

Reading the band
Each bar sits on the weeks its own work is named in. The week 5 overlap is worked time, not padding.
At the end of W6
Once the wet season validates, Agent Care assumes the running agent.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Construction AI agent

Build a stormwater agent around the permit each site holds.

Show us your permit coverages, your inspection reports and the two people who sign. What actually gets cited? A report missing an element its own permit lists — EPA priced exactly that against Washington state permit conditions in its 2025 expedited settlements.

Typical delivery
4–6 weeks
Deployment
Your permits and feeds
Two signatures
Pre-submission
After launch
Agent Care available

Nestack Agents · Stormwater compliance

AGT-CN-10 · Agent Care available after launch

Build a stormwater agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your sites

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.