Disclaimer: Independent product / portfolio work by Kaushal Khodifad. Not a live commercial product.return to portfolio
Disclaimer: Independent product / portfolio work by Kaushal Khodifad.
A Spotter AI · Sentinel teardownFMCSA-grounded · 49 CFR 391/382Synthetic data - no real CDLs / PII

Sentinel hands you a black-box safety score. Spotter hands you the decision - and the receipts.

Spotter AI’s Sentinel pulls MVR, PSP, FMCSA and CDLIS on a CDL upload in ~30 seconds and returns an AI safety score. This is a teardown of that pattern: Paste a synthetic CDL and watch a real, auditable engine assemble a driver risk file - every point of risk traced to a specific FMCSA violation and the CFR behind it, with an auto-approve / manual-review / deny verdict you could defend to an auditor.

The wedge

Underwriting trust, not data access, is the moat.

01

Glass-box FMCSA-grounded scoring - replicate the real SMS construction so every point is explainable line-by-line, not a black box.

02

Decision, not just data - an explicit auto-approve / manual-review / deny verdict with a reason ledger tied to specific violations and the governing CFR.

03

Counterfactual transparency - toggle any signal and the file is re-underwritten from scratch; every delta is verified against an independent spec recompute, and a built-in fuzz harness shows the checks failing against deliberately broken engine variants.

04

Grounded incident triage - an LLM that retrieves the driver record + carrier policy and produces a cited RCA that abstains on thin evidence.

The teardown · same driver, two products

A black-box score vs. a reconstructable reason ledger

SambaSafety and Spotter AI’s Sentinel hand a carrier a number; the same synthetic driver, run through this engine, returns the number and the line-by-line derivation - each point traced to a violation and its governing CFR.

Incumbent - black-box scoreSentinel / SambaSafety pattern
100
composite risk score (0-100)
~75-95th pct
The carrier sees a score and a verdict. Why it is what it is - which violation, how old, how it was weighted, which rule it maps to - is not exposed. There is nothing to defend to an auditor and nothing to recompute.
  • No per-violation point derivation
  • No citable CFR behind the number
  • No counterfactual - can't see what each signal added
Spotter - glass-box reason ledgerdriver Driver B-2207
100
same index = 74 pts ÷ 6 exposure × 10
verdict: DENY
ViolationCFRWhenPts
392.2-SLLS2
Speeding 6-10 mph over the limit
49 CFR 392.23 mo ago15
392.2FC
Following too close
49 CFR 392.25 mo ago15
392.2FC
Following too close
49 CFR 392.25 mo ago15
393.47
Brakes out of adjustment
49 CFR 393.4711 mo ago12
392.82
Using a hand-held mobile phone while driving
49 CFR 392.829 mo ago10
Top 5 of 7 scored line items shown. Each point = severity × time-weight; toggle any line in the Workbench and the file is re-underwritten from scratch, with the delta verified against an independent spec recompute (under a capped inspection it is not simply that line’s points - the cap backfills).
Open the full ledger →

Illustrative comparison on synthetic driver Driver B-2207. SambaSafety and Spotter AI are real companies; the score, line items, and verdict here are produced by this prototype’s engine on synthetic data, not by those vendors’ systems.

The pipeline · five stages to explore
Click into any stage →
How one point is computed
deterministic · reconstructable
Violation
1 inspection
Severity
legacy 1-10 / 2025 1-2
× Time-weight
3 / 2 / 1
Cap @ 30
worst-first, per BASIC
Point
line item
Σ Sum
all points
÷ Exposure
synthetic divisor
Index 0-100
× 10
11
synthetic driver personas
35
FMCSA-style violation codes
3 / 2 / 1
real SMS time weights
2
severity scales (legacy + Dec-2025)
< 16 ms
deterministic recompute
The CUO math

Who pays, and what a defensible verdict is worth

A worked example, not a business case: every row is badged for what it is, and every derived number is plain arithmetic over the badged inputs. Nothing below is a measured outcome of this prototype.

Inputs (badged)
Fleet size (worked example)A mid-size truckload carrier; scale linearly for your fleet.500 power unitsillustrative
Annualized driver turnoverATA: 90-95% for large truckload carriers (~92.7% long-run avg).90%verified
Driver files underwritten / yr500 seats × 90% turnover.450arithmetic
Manual review time per fileRecruiter + safety-manager time across CDLIS/MVR/PSP/Clearinghouse review.45 minillustrative
Files a glass-box verdict auto-approvesClean files with no gate, no recent OOS, index below the review band.60%illustrative
What falls out (arithmetic only)
Files removed from the manual queue450 files × 60% auto-approve.270 / yrarithmetic
Safety-team hours returned270 files × 45 min.~202 hrs / yrarithmetic
Budget headroom from ONE avoided average crash$91K avg crash cost (FMCSA 2025, verified) ÷ 450 files - the willingness-to-pay anchor a CUO prices against.~$202 per filearithmetic
Downside of ONE bad auto-approveInjury ($200K) to fatality ($3.6M) crash cost - why the deny/review gates are hard-coded law, not tunable knobs.~$200K-$3.6Mverified

Pricing hypothesis (directional): per driver-file underwritten, anchored to the screening spend Sentinel claims to cut ~75% (vendor claim) and to the ~$202/file headroom one avoided crash creates. Not a quote; no unit economics are measured by this prototype. Sources and audit status for the verified figures are on the Methodology page.

Audit status · every load-bearing number is sourced
See the full ledger →
9
CFR citations in the engine
3
market stats fact-verified
3
broken variants fuzzed against
0
tautological checks (v1 bug fixed)
What is real here, and what is simulated
Real
  • Deterministic engine + ledger arithmetic
  • Falsifiable check suite + fuzz harness
  • OpenRouter LLM triage - labeled live vs. fallback per run
Public-sourced
  • SMS mechanics: time weights, +2 OOS, cap-30
  • Dec-2025 severity overhaul
  • 49 CFR 382/391 hard gates
Synthetic
  • 11 driver personas + violation histories
  • Severity table values (representative)
  • Exposure divisor + 0-100 index map + peer bands

↑ Independent product by Kaushal Khodifad. Spotter AI is a personal project / portfolio piece in the commercial driver onboarding & fleet safety space. Not a live commercial product. Data is illustrative.

Return to portfolioOpen the live demo