Automotive AI agent · EV high voltage
EV High-Voltage Service-Authorisation AI Agent
Check that the technician booked onto a pack job is qualified and current, that the gloves and meter are in test and rated, and hold the job for the shop manager to authorise.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Manager signs
- Pre-work check
- After launch
- Agent Care
What this agent does
Checks the job, never authorises it
In
- 01
A pack job is booked, and the vehicle, the task and the pack architecture are read from the shop system.
- 02
A technician is assigned, and the credential on file is matched to the task, not to the ticket wording.
Reason
- 03
A pack is opened by a qualified person — outside Subpart S by 1910.331(c)(3), so Subpart I and 1910.147 carry it.
- 04
A glove test falls due at six months under Table I-5, and the pair, not the technician, holds that clock.
- 05
A boxed pair tested over twelve months ago is kept out of service, even though it was never issued.
Decide
- 06
A meter is read for its CAT III or CAT IV rating against the pack voltage, not only for a serial number.
- 07
An isolation verification is sought per de-energisation — that record is an OEM or shop artefact, not OSHA's.
Out
- 08
A state-of-health read is kept with its interface and date — EU Article 14 access dates from 18 August 2024.
- 09
A clearance or a block goes to a named person, and any write-back stays inside the agreed boundaries.
Product statement
The agent checks currency and completeness; the shop manager authorises the work, and no agent may authorise work on a live high-voltage system.
Example workflow
One pack job, booking to authorisation
- 1Pack job receivedRepair order, VIN, task list, pack architecture and the technician assigned
- 2Currency assembledCredential level and expiry, glove and blanket test dates, meter rating and calibration
- 3Checks runCredential, PPE, instrument, isolation record and confidence
- 4Controls appliedLevel-to-task checks, 1910.137 Table I-5 intervals, instrument-rating checks and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is authorised at any of them — the agent is checking, and the manager's lane opens at the confidence gate.
Goes to the shop manager to authorise.
Adds a safety-lead read first.
Manager authorisation
The job is held with its checks, its failed items and the confidence.
Authorise · Send back · Escalate to safety lead
Authorised — released to the bay
- 6Shop systems updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedCredential currency, PPE test lag, blocked jobs and findings raised after the job
Blocks
Every blocked job is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Declaring a high-voltage system de-energised.
- Signing the absence-of-voltage verification.
- Deciding that a technician is qualified.
- Valuing a pack or calling repair against replace.
Automation boundary
Agent acts unaided
- ✓Check the credential level and expiry against the task booked.
- ✓Test glove, blanket and instrument dates against.
- ✓Check the instrument category rating against the pack.
- ✓Name what is missing and hold the job for the shop manager.
Any write-back runs inside the boundaries agreed at implementation, never ahead of authorisation.
- Overriding a failed check on the shop floor.
- Setting the shop's re-qualification interval.
- Whether 13 CCR 1962.5 binds this vehicle.
- Changes to credential, PPE or clock rules.
Example output
One pack job, annotated
Everything the agent checks is attached to the job it was read from.
Authorisation check · single job
Illustrative example- Vehicle
- Crossover, liquid-cooled pack
- Task booked
- Pack opened for a module replacement
- Architecture
- Above 600 V nominal
- Credential of record
- xEV Level 2, in date
- Confidence
- 91%
- Outcome
- Held for the shop manager
As received
Taken from the repair order, the training log and the glove register — nothing on this side is decided by the agent.
Checks assembled
- Training log entry
- Glove test register
- Meter calibration
Why the job is held
The pack sits above 600 volts nominal, so shop policy proves the meter after the test as well.
What the score decides
Below the configured threshold the job picks up a safety-lead read before it.
Action
AuthoriseSend backEscalate to safety leadValue
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 high-voltage job
From the repair order
Check the person against the task
Read the training log, the credential expiry and the shop's own re-qualification interval.
Qualified is the employer’s word
A certificate is an input to that call — ASE's xEV standards v2.1 leave the duty with the employer.
Put the read where a check failed
Jobs with a lapsed credential, an out-of-test pair or an unrated meter are marked, so the manager starts there.
Build an evidence trail
The task, the credential it was checked against and the technician who verified stay on the vehicle.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Shop management
- Tekmetric
- Shop-Ware
- R.O. Writer
- CDK
- Reynolds
Training and credentials
- ASE credential records
- OEM training portals
- LMS
Tool and PPE registers
- Glove test certificates
- Meter calibration records
Agent
EV high-voltage authorisation
- Reads the job
- Checks currency
- Holds for authorisation
Diagnostics and state of health
- J1962
- SAE J1979-3
- OEM scan tools
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 bay
Six stages of check, outermost first. What no stage catches is written into the map beneath it.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether the job may proceed
- L3
- Puts the authorisation in a manager's hands
- L4 – L5
- Keep the task and the credential behind it
- L6
- Holds the vehicle at unauthorised when signals degrade
How Nestack evaluates it
Evaluate the whole gate — not only the credential lookup.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the authorisation the shop stands behind
- E1Final-output evaluationDid the check match the task actually booked?
- E2Step-level evaluationDid the agent read the right log, register and shop policy?
- E3Tool evaluationDid it read the right technician and the right pair of gloves?
- E4Confidence calibrationDo low-confidence checks actually get overturned more?
- E5Slice evaluationHow does performance change across specific technicians?
- E6Business outcomeHow many jobs were authorised and later found short on a check?
Floor — the outcome the shop answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, placed at the stage each one originates.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- HV-03
Stale credential source
Expiry read from the issue date, not the test date.
Stage gathers
Repair order, training log, PPE and tool registers
02 · Reasoning
2 modes
- HV-04
Level matched to wording
A task list saying disconnect battery hides an MSD.
- HV-06
Rating never read
PPE is logged as present but not as rated and in test.
Stage proposes
Credential, PPE, meter rating and confidence
03 · Tool / write
2 modes
- HV-02
Clearance without a name
A job clears with no named person behind it.
- HV-05
Isolation recorded once
A second de-energisation reuses the first record.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- HV-01
Certificate taken as proof
The employer's verification of ability is missing.
Stage returns
The clearance a manager turns into authorisation
05 · Change / Version
1 mode
- HV-07
Silent clock regression
A rule change lengthens an interval the agent tests.
Stage tracks
Model, prompt, clock rules and credential sources
- Sev-1 · a job starts without authorisation
- Sev-2 · a lapsed credential clears a check
- Sev-3 · source degrades, job routes to review
Affected slices
A handful of technicians carry the lapses
A shop-level currency figure can look healthy while a few technicians carry most of the lapsed checks. Nestack reports the failed-check rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Contract and mobile technicians | 6.1% | 3.5× | Review | |
| Shared glove pairs | 3.9% | 2.2× | Review | |
| Recently certified technicians | 2.7% | 1.5× | Watch | |
| Named in-house technicians | 1.5% | 0.9× | Normal |
Bar: failed-check lift vs. in-house-technician baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
A cycle ends with one more case
A cycle is done when the lapsed credential is a case the next release must close. That suite is what the next authorisation given is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Failed-check rate rises in one technician group.
- 02
Diagnose
The tech who opened a pack on a certificate that expired in spring is traced back to the clock the agent read.
- 03
Improve
Version the fix, and the vehicles that exposed it stay attached to that version.
- 04
Verify
The release is held back until each touched credential case passes again.
- 05
Learn
The case is retained, and the currency rules move with it in one change.
Learn → Detect
The return edge. The next lapse is looked for 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, records, check workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01High-voltage workflow discovery and boundary setting.Week 1
- 02Training-log and tool-register assessment.Week 1, Week 2
- 03Credential, PPE and instrument-clock rule mapping.Week 1, Week 2
- 04Job and technician ingestion.Week 2
- 05Currency checks and level-to-task logic.Week 2, Week 3
- 06Confidence scoring and block routing.Week 3
- 07Manager authorisation workflow.Week 3, Week 4
- 08Shop and training-record integration.Week 5
- 09Credential and currency cases.Week 4, Week 5
- 10Guardrails and authorisation controls.Week 4
- 11Vehicle-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 site, one shop | ProductionProduction shop systems | AdvancedMultiple sites / brands |
|---|---|---|---|
| Introduced at Pilot | |||
| Currency checks on your records | ✓ | ✓ | ✓ |
| Manager authorisation | ✓ | ✓ | ✓ |
| Credential-currency baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by technician | — | ✓ | ✓ |
| Authorisation workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Training-record integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-site clock rules | — | — | ✓ |
| Multi-stage shop approvals | — | — | ✓ |
| High job volume | — | — | ✓ |
| Multi-site authorisation 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 technician records and training log
→Job and technician ingestion and credential mapping
Week 102Representative closed high-voltage jobs
→Currency baseline, clock mapping and check binding
Week 203Your PPE register and tool inventory
→Credential, PPE and instrument-clock rule mapping
Week 104Access to relevant APIs, feeds or exports
→Training-log and tool-register assessment, then integration setup
Week 205Work you would not want authorised
→Isolation cases and failure-mode testing
Week 406What no authorisation may skip
→Confidence scoring, block routing, guardrails and authorisation controls
Week 307A named shop manager and safety lead to review
→Manager authorisation 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
Each band covers the weeks the rollout really takes, so evaluation and pilot share the fifth.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1High-voltage workflow discovery, clock mapping and the boundary
W2Record integration and the currency baseline
W3Check logic, confidence scoring and authorisation controls
W4Evaluation suite, isolation-record checks and failure-mode testing
W5Training-record integration, pilot jobs and targeted corrections
W6One service cycle run under the shop manager, 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
- Validation closes on live vehicles and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Automotive AI agent
Build an authorisation gate around your shop's own qualification record.
Show us your training log, your glove register and who authorises a pack job. That record has to survive an inspection under 1910.132 and 1910.147, an insurer's file review, and a question asked years later.
- Typical delivery
- 4–6 weeks
- Deployment
- Your shop records
- Manager signs
- Pre-work check
- After launch
- Agent Care available
Nestack Agents · EV high-voltage service authorisation
AGT-AUT-17 · Agent Care available after launch