Transportation AI agent · Maintenance
Fleet-Maintenance Planner AI Agent
Schedule maintenance from condition signals and your own intervals, report the precision and lead time behind each alert, and leave certifying, releasing and clearing to a qualified inspector.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Inspector signs
- Pre-inspection
- After launch
- Agent Care
What this agent does
Schedules the work, never certifies it
In
- 01
Ingest condition signals, fault codes, DVIR entries and service intervals from supported maintenance sources.
- 02
Normalise unit, component and fault fields, and carry each reading forward with the source it came from.
Reason
- 03
Rank units against your own intervals and the open defects already reported on them.
- 04
Raise an alert with the precision it has on your data and the lead time it bought, per component.
- 05
Say when a signal is thin, because a rarer failure gives the same model a worse precision.
Decide
- 06
Hold the periodic inspection on its own schedule — a prediction adds work, it never subtracts it.
- 07
Route the job to the shop, and every certifying act to a qualified inspector.
Out
- 08
Retain the signal, the interval, the decision taken and the decision not taken, against the unit.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The planner schedules; a qualified inspector under 49 CFR 396.19 signs, and the carrier keeps the systematic duty under 396.3.
Example workflow
One unit, signal to shop
- 1Condition signal receivedTelematics feed, fault code, DVIR entry or service record
- 2History and intervals gatheredUnit history, service intervals, open DVIR defects and component signals, each with its source
- 3Work proposedJob, priority, alert precision and lead time
- 4Controls appliedInterval checks, open-defect holds, out-of-service holds and confidence threshold
No human action required
Stages 1 to 4 run unaided and certify nothing — scheduling is not signing, and the inspector's lane opens at the confidence gate.
Goes to the shop to schedule.
Adds a planner's read first.
Inspector signature
The job is held with its signal, the interval behind it and the confidence.
Schedule · Adjust · Send to the shop
Signed — released by a person
- 6Maintenance systems updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedAlerts raised, no-defect-found visits, lead time gained and what the shop actually found
Adjustments
Every shop adjustment is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Certifying a repair or a periodic inspection.
- Releasing a unit back into service after repair.
- Clearing or closing a defect reported on a DVIR.
- Lifting an out-of-service marking from a unit.
Automation boundary
Agent acts unaided
- ✓Rank units against your own intervals and their signals.
- ✓Report the alert precision and lead time on your own data for the named owner.
- ✓Book the work into the shop diary and the unit's file.
- ✓Block dispatch on an open defect and hold the unit for the shop.
Any write happens inside the boundaries agreed at implementation, never ahead of approval.
- Deferring a periodic inspection on a prediction.
- Judging whether a brake inspector is qualified.
- Deciding a unit is roadworthy for the next run.
- Changes to the interval set or alert thresholds.
Example output
One unit, annotated
Everything the planner proposes is attached to the signal it was raised from.
Maintenance output · single unit
Illustrative example- Unit
- Day-cab tractor
- Proposed job
- Air-pressure trend outside the configured band
- Lead time
- 9 days of lead
- Signal source
- Telematics air-system channel
- Confidence
- 86%
- Who signs
- A qualified inspector under 396.19
As received
Taken from the telematics channel and the unit's service file — nothing on this side is predicted by the agent.
Signals used
- Air-system channel trend
- Prior work on this unit
- Open DVIR entries
Why it was raised
It is an inspection to book, not a fault found — the shop decides what is there.
What the score decides
Below the configured threshold the job picks up a planner's read before the shop sees it.
Action
ScheduleAdjustSend to the shopValue
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 unit
From telematics and the service file
Schedule against your intervals
Draw on your own service intervals, the unit's history and the condition signals you already collect.
Schedule it, do not certify it
Routine interval work and inspection due-dates arrive already booked.
Hand the rest to the shop
Thin signals, open defects and low-confidence jobs are marked, so the shop's read starts where risk concentrates.
Build an evidence trail
The condition signal, the interval behind it and the mechanic who signed it stay on the unit.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Maintenance systems
- TMT Fleet
- Dossier
- Fleetio
- Whip Around
Telematics and diagnostics
- Samsara
- Motive
- Geotab
- Noregon
Parts and shop
- Parts catalogues
- vendors
- Shop diaries
- work orders
Agent
Fleet-maintenance planning
- Reads the signal
- Schedules the work
- Holds for the inspector
Dispatch and inspection
- TMS
- dispatch boards
- DVIR
- inspection 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 unit
Each control encloses the last. Whatever slips past all of them is named below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a job may stand
- L3
- Puts the signature in an inspector's hands
- L4 – L5
- Keep the signal and the interval behind it
- L6
- Cuts back to interval scheduling when signals degrade
How Nestack evaluates it
Evaluate the whole planning workflow — not only the alert at the end.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the job the shop receives
- E1Final-output evaluationDid the shop find what the alert said was there?
- E2Step-level evaluationDid the planner use the right intervals, history and signal channels?
- E3Tool evaluationDid it read the correct unit and write the correct work order?
- E4Confidence calibrationDo low-confidence jobs actually end in no defect found?
- E5Slice evaluationHow does precision change across specific component families?
- E6Business outcomeHow many alerts were dismissed, and how much lead time did the rest buy?
Floor — the unit that has to pass inspection
Failure modes
Where each failure originates in the agent
Seven ways a schedule goes wrong, at its own stage.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- AM-03
Stale service record
The interval is read from a record the shop has since updated.
Stage gathers
Unit history, intervals, open defects and signals
02 · Reasoning
2 modes
- AM-04
Alert without precision
A component is flagged with no precision behind the flag.
- AM-06
Interval quietly deferred
A periodic inspection slips because a model called it healthy.
Stage proposes
The component, the priority and the lead time
03 · Tool / write
2 modes
- AM-02
Job released without a person
A write books or closes work before anyone has signed.
- AM-05
Repeat alert on one unit
The same signal raises a job the shop already dismissed.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- AM-01
Open defect not held
An open DVIR defect does not stop the unit being scheduled out.
Stage returns
The job the shop reads and the unit it books
05 · Change / Version
1 mode
- AM-07
Silent threshold regression
A model or rule change quietly widens what raises an alert.
Stage tracks
Model, prompt, interval rules and thresholds
- Sev-1 · a unit is released without a signature
- Sev-2 · a required inspection is missed
- Sev-3 · a signal degrades, job routes to review
Affected slices
One pooled precision hides the rarer component
Precision moves with the base rate, not only with the model, so a family whose failures are rarer returns more no-defect-found visits from the same alert engine. Nestack reports that rate by component family, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Aftertreatment and DPF signals | 7.1% | 3.3× | Review | |
| Air-system and brake wear | 6.2% | 2.9× | Review | |
| Battery and charging signals | 4.3% | 2.0× | Watch | |
| Air-pressure (APS) signals | 1.9% | 0.9× | Normal |
Bar: no-defect-found-rate lift vs. air-pressure-signal baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Close it with a case, not a theory
The cycle does not end in a post-mortem. It ends in a case the next release has to clear, and that suite is what the next unit scheduled is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
No-defect-found rate rises in a component family.
- 02
Diagnose
Read the alerts against what the shop actually found, until one cause holds.
- 03
Improve
Each change carries a version and the units that forced it.
- 04
Verify
Release waits on the affected cases going green again.
- 05
Learn
The case is kept for good, and the maintenance programme is rewritten.
Learn → Detect
The return edge. The next detection runs against 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, sources, planning workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Maintenance workflow discovery and boundary definition.Week 1
- 02Telematics and shop-system assessment.Week 1, Week 2
- 03Service interval, defect-hold and threshold mapping.Week 1, Week 2
- 04Signal and service-record ingestion.Week 2
- 05Scheduling logic and interval binding.Week 2, Week 3
- 06Confidence scoring and defect routing.Week 3
- 07Inspector signature workflow.Week 3, Week 4
- 08Maintenance and shop-system integration.Week 5
- 09Defect and interval cases.Week 4, Week 5
- 10Guardrails and out-of-service controls.Week 4
- 11Unit-trail instrumentation.Week 5, Week 6
- 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 tier | PilotOne terminal, one unit class | ProductionProduction maintenance systems | AdvancedMultiple terminals / classes |
|---|---|---|---|
| Introduced at Pilot | |||
| Scheduling to your own intervals | ✓ | ✓ | ✓ |
| Inspector signature | ✓ | ✓ | ✓ |
| Schedule-quality baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by unit class | — | ✓ | ✓ |
| Signature workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Maintenance-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-class interval sets | — | — | ✓ |
| Multi-stage shop approvals | — | — | ✓ |
| High unit volume | — | — | ✓ |
| Multi-terminal fleet controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction volume, 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.
You bring
We build with it
01Your service intervals and unit classes
→Signal and service-record ingestion and interval mapping
Week 102Representative historical work orders
→Scheduling baseline, interval binding and per-component precision
Week 203Your defect holds and alert thresholds
→Service interval, defect-hold and threshold mapping
Week 104Access to relevant APIs, feeds or exports
→Telematics and shop-system assessment, then integration setup
Week 205Units you would not want released
→Defect cases and failure-mode testing
Week 406What must reach a qualified inspector before a unit rolls
→Confidence scoring, defect routing, guardrails and out-of-service controls
Week 307Named qualified inspectors to sign
→Inspector signature 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 bands sit on the weeks the work occupies, which is why the fifth carries two kinds at once.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Maintenance workflow discovery, interval mapping and the boundary
W2Telematics and shop-system integration and the scheduling baseline
W3Planning workflow, confidence logic and signature controls
W4Evaluation suite, defect cases and failure-mode testing
W5Shop-system integration, a pilot unit class and targeted corrections
W6One maintenance cycle planned under the shop, then Agent Care handover
- Reading the band
- Each bar covers only the weeks its work is named in. The week 5 overlap is real, not padding.
- At the end of W6
- The cycle closes the validation and Agent Care takes over monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Transportation AI agent
Build a maintenance planner that schedules and leaves the signature alone.
Show us your intervals, your signals and your shop. Then answer one question: when a model calls a unit healthy, who signs that the inspection can wait? We set the boundary around that answer.
- Typical delivery
- 4–6 weeks
- Deployment
- Your maintenance stack
- Inspector signs
- Pre-inspection
- After launch
- Agent Care available
Nestack Agents · Fleet maintenance
AGT-TR-05 · Agent Care available after launch