Transportation AI agent · Driver qualification
Driver Qualification & Clearinghouse AI Agent
Reconcile each driver qualification file against what 49 CFR 391.51(b) actually names, run every driver's own twelve-month clock, and hold the file for the person who reviews and dates it.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Consent first
- Name and date
- After launch
- Agent Care
What this agent does
Assembles the file, never the qualification
In
- 01
A driving record is requested, and § 391.25(a) puts that inquiry at least once every 12 months.
- 02
A record comes back reviewed, and § 391.25(c) wants the reviewer name and the review date in the file.
Reason
- 03
A twelve-month clock falls due per driver, and § 382.701(b)(1) dates it from the last query, not January.
- 04
A limited query returns information, and § 382.701(b)(3) allows 24 hours for the full query.
- 05
A full query is next, and § 382.703 wants the driver's own submission on top of the consent you hold.
Decide
- 06
A medical certificate is filed, and § 391.51(b)(8) reads Through June 22, 2025 for CDL and CLP holders.
- 07
A safety performance history lands, and § 391.23(c)(1) files it within 30 days of employment start.
Out
- 08
A return-to-duty negative arrives, and § 382.705(b) reports it by close of the third business day.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent assembles and tracks each clock. A named person performs the review and dates it; only an employer notified in January submits a summary report.
Example workflow
One driver file, document to review
- 1Document or query result receivedLicensing records, medical certifications, applications, road tests and query returns
- 2File context assembledThe driver, the class held, the section each document answers to and the date it carries
- 3File state draftedThe documents present, their versions, the gaps and completeness
- 4Controls appliedConsent checks, clock checks, document-currency checks and completeness confidence
No human action required
Stages 1 to 4 run unaided, and no driver is qualified at any of them — the agent is assembling, and the safety lane opens at the completeness gate.
Goes to the named person to review and date.
Adds a safety-desk read first.
Safety review
The file is held with its gaps, its clocks and the driver class it sits in.
Review · Append document · Send to the safety desk
Reviewed — by the named person, dated
- 6Driver and training records updatedOnly where write access and records policy allow it
- 7Outcome evaluatedDocument currency, clock coverage, desk corrections and what review found
Corrections
Each safety-desk correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Determining that a driver is qualified to drive.
- Performing the annual review and dating the note.
- Certifying the summary report a company official signs.
- Querying the Clearinghouse without driver consent.
Automation boundary
Agent acts unaided
- ✓Hold every document § 391.51(b) names, with the date.
- ✓Track each driver's own twelve-month clock from the last query made.
- ✓Chase a previous employer who never replied and log it.
- ✓Flag the licensing record where a paper certificate is still on file.
Nothing is reviewed or certified except by a named person, inside agreed boundaries.
- Deciding whether a driver may keep working.
- Telling an auditor a driver file is complete.
- Setting the driver class a file is kept at.
- Changes to driver, training or licensing records.
Example output
One driver file, annotated
Our driver support copilot stops the moment a driver cites hours, fatigue or a defect; this record is what one qualification file held.
File state · single driver
Illustrative example- Document
- Annual review note, § 391.25(c)
- Recorded as
- Reviewer name recorded, review dated
- Driver class
- CDL holder
- Source of record
- Licensing authority record, 3 August 2026
- Confidence
- Held unreviewed
- Held for
- The reviewer, by name
As received
Taken from the licensing authority record and the file itself — it reaches as far as those sources do.
What the file holds
- Licensing record date
- Annual inquiry return
- Review note, dated
Why no qualification call here
Whether a driver is qualified is a § 391.25(b) judgement, not a model output.
What the score decides
Below the configured threshold the file picks up a desk read before the reviewer sees it.
Action
ReviewAppend documentSend to the safety deskValue
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 document
From the source that issues it
Where the document is used
Our safety coaching agent keeps its score out of hiring, discipline and discharge — that is right, and this file is what hiring actually runs on.
A name and a date, not a form
§ 391.25(c) wants the reviewer name and the date of review in the file, and asks for no signature.
What was checked, and not found
An ELDT-II Final Rule is referenced on the agency's own page and no matching Federal Register document was found; the verifiable 2025 action is the Training Provider Registry purge, with providers removed and others given 30 days to answer.
Build an evidence trail
The document, the version it is and the reviewer who dated it 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.
Licensing and medical records
- CDLIS motor vehicle record
- National Registry verification
Clearinghouse queries
- FMCSA Clearinghouse portal
- Query and consent records
Driver and HR records
- Driver files
- HR systems
- Applications and road tests
Agent
Driver qualification and Clearinghouse
- Reads the file
- Tracks each clock
- Holds for the reviewer
Training and testing
- Training Provider Registry
- Drug and alcohol test records
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six meshes between the model and the reviewer
Six meshes, the coarsest leading. Whatever drops through every one is set out in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a file may stand
- L3
- Puts the review in a person's hands
- L4 – L5
- Keep the document and the version behind it
- L6
- Holds the file incomplete when signals degrade
How Nestack evaluates it
Evaluate the whole assembly — not only the file that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the file an auditor reads
- E1Final-output evaluationDid the file hold what each section actually asks for?
- E2Step-level evaluationDid the agent read the right driver, the right source and the current record?
- E3Tool evaluationDid it read and write the correct driver file and the correct query?
- E4Confidence calibrationDo low-confidence files actually attract more desk corrections?
- E5Slice evaluationHow does performance change across specific driver classes?
- E6Business outcomeHow many files needed a correction before the reviewer dated one?
Floor — the file the carrier answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each placed at the stage where it first shows.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- JE-03
Stale licensing record read
The record read is not the one now on the licence.
Stage gathers
The drivers, the documents, the clocks and the dates
02 · Reasoning
2 modes
- JE-04
File asserted, not shown
A file is called complete without its documents.
- JE-06
Proposed rule read as live
The language proficiency NPRM is worked as law.
Stage proposes
The documents, their versions and completeness
03 · Tool / write
2 modes
- JE-02
Thin file passed forward
A file moves on without the safety-desk read.
- JE-05
Document bound to wrong driver
A record is filed against the wrong driver.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- JE-01
Reviewed, name unrecorded
The file shows a review but not who performed it.
Stage returns
The file a reviewer dates and an auditor reads
05 · Change / Version
1 mode
- JE-07
Silent clock regression
A configuration change moves the clock, not the file.
Stage tracks
Model, prompt, file rules and document fields
- Sev-1 · a file complete on no documents
- Sev-2 · wrong document reaches the file
- Sev-3 · source degrades, file holds unreviewed
Affected slices
Non-domiciled CDL holders absorb the corrections
A class-level document-currency figure can read clean while non-domiciled CDL holders carry most of the corrections. Nestack reports the correction rate by driver class, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Non-domiciled CDL holders | 10.3% | 3.7× | Review | |
| New hires inside 30 days | 7.7% | 2.7× | Review | |
| Drivers on a mid-year clock | 4.9% | 1.7× | Watch | |
| Long-tenured CDL holders | 1.8% | 0.6× | Normal |
Bar: correction-rate lift vs. long-tenured CDL baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What a superseded document costs
A cycle closes when the superseded artefact is a regression case. That suite is what the next file assembled is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on non-domiciled CDL holders.
- 02
Diagnose
The medical certificate somebody filed on paper for a driver whose licence already carried it is read back until one cause remains.
- 03
Improve
The change ships numbered, and the files that forced it ride with it.
- 04
Verify
Nothing releases while one touched driver case is still red.
- 05
Learn
It is retained for good, and the file rules are amended in that same commit.
Learn → Detect
The return edge. The next file is measured 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, file assembly, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Driver-file discovery and automation-boundary work.Week 1
- 02Licensing, medical and query sources.Week 1, Week 2
- 03Document-to-section and clock-coverage mapping.Week 1, Week 2
- 04Driver document ingestion.Week 2
- 05Driver, class and source binding.Week 2, Week 3
- 06Completeness scoring and review routing.Week 3
- 07Named-reviewer review workflow.Week 3, Week 4
- 08Driver and licensing-system integration.Week 5
- 09Document-currency and query cases.Week 4, Week 5
- 10Guardrails and review controls.Week 4
- 11File-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 class, one review year | ProductionProduction driver-file workflow | AdvancedMultiple classes / terminals |
|---|---|---|---|
| Introduced at Pilot | |||
| File assembly to your sections | ✓ | ✓ | ✓ |
| Named-reviewer release | ✓ | ✓ | ✓ |
| Document-inventory baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by driver class | — | ✓ | ✓ |
| Review workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Licensing-record integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-terminal rosters | — | — | ✓ |
| Cross-section document packs | — | — | ✓ |
| Large driver rosters | — | — | ✓ |
| Multi-source consent controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, roster size, 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 drivers and the classes they hold
→Roster mapping and document capture
Week 102Representative licensing, medical and query records
→Source binding, clock logic and the document baseline
Week 203Your consents held under § 382.703
→Document mapping, driver binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Licensing, medical and query-source assessment, then integration setup
Week 205Files you would not want reviewed
→Currency cases and failure-mode testing
Week 406What no qualification file may establish
→Completeness scoring, review routing, guardrails and release controls
Week 307A named person to perform the review
→Named-reviewer 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
These bands are the weeks each phase honestly costs, and the fifth of them therefore carries a pair.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Driver-file discovery, document mapping and the automation boundary
W2Source integration and the document-currency baseline
W3File assembly, clock logic and release controls
W4Evaluation suite, currency cases and failure-mode testing
W5Record integration, pilot files and targeted corrections
W6One review year run under the safety director, then Agent Care handover
- Reading the band
- Each bar covers only the weeks its own work is named for. The fifth week carries two because the work does.
- At the end of W6
- Validation closes on live files, and Agent Care picks up the watch.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Transportation AI agent
Build a driver qualification agent around the file your named reviewer has to date.
Show us one driver file and the documents behind it. If a CDL holder on your roster still has a paper medical card in the file, then it is running on a template that stopped matching § 391.51(b)(8) on 23 June 2025. Another carrier's authority at tender is a different agent.
- Typical delivery
- 4–6 weeks
- Deployment
- Your driver roster
- Consent first
- Name and date
- After launch
- Agent Care available
Nestack Agents · Driver qualification
AGT-TR-18 · Agent Care available after launch