Levy Vision FAQ

This article answers the questions we hear most about Levy Vision. If your question isn't here, check Troubleshooting or contact support.

Accuracy & performance

How accurate is the helmet check?

The classifier targets 92%+ true-positive rate (correctly identifying a helmet when one is present) and under 5% false-positive rate (wrongly identifying a hat or hood as a helmet). Real-world numbers from our pilots have landed in those ranges, with weekly improvements as we feed back labeled false-positive flags.

How accurate is sidewalk detection?

Vendor-dependent — Drover and Luna both publish 85–90% precision numbers from their fleet partners. Levy Vision tunes conservative on top, so the policy ladder doesn't fire until events repeat within a single ride. That keeps the effective false-positive rate down at the rider-facing layer.

How accurate is parking-pose?

The Gemini-based classifier targets 90%+ upright-detection accuracy. Captur.ai (premium) is a few points higher on edge cases. Most operators don't see a meaningful improvement going from Gemini to Captur unless they have a high-volume fleet where the marginal accuracy is worth the cost.

What's the latency from photo to result?

• Helmet: 2 seconds at p50, 4 seconds at p95
• Parking (Gemini): 3 seconds at p50, 6 seconds at p95
• Parking (Captur): async — the rider doesn't wait, the result lands via webhook
• Sidewalk (vendor): 2–4 seconds from on-vehicle detection to rider warning

Privacy & legal

Do you store rider faces as biometric data?

No. Helmet selfies are stored as images for evidence and model improvement, but we do not extract face embeddings, run facial recognition, or build any biometric template. The classifier asks "is there a helmet in this photo?" — that's it.

How long do you keep selfies and parking photos?

By default, 60 months. This is the spec retention for insurance defensibility. You can lower this per subaccount via the cv_evidence_retention_days setting if your jurisdiction requires shorter retention. Contact support to adjust.

What about Illinois BIPA?

Riders billed to Illinois see an explicit BIPA consent toggle during onboarding before any selfie is requested. If they don't consent, no selfie is captured and helmet enforcement is skipped for that rider. The toggle is opt-in, not opt-out.

What about GDPR and CCPA?

Both. The lawful basis is legitimate interest (safety) and contractual necessity (the rental agreement). Riders can request deletion via the standard data-deletion flow, which purges their CV evidence along with other personal data.

Can my city see rider faces?

No. The MDS export strips all image URLs and rider identity before leaving the system. Cities get event metadata (timestamp, location, type, confidence) but never images or identifiable rider info.

Pricing & packaging

Is Levy Vision included in my plan?

Phase 0 (helmet + parking-pose with Gemini) is included in Pro and Enterprise tiers at no extra cost. Free tier is capped at 200 helmet checks per month. There's no per-event charge for the included features.

How much do Drover and Luna cost?

Approximately $300 per device upfront and $5–15 per device per month, billed by the vendor directly. These ranges are estimates from public reporting; your actual quote depends on volume and vendor negotiation. We don't mark up vendor hardware — you buy direct.

How much does Captur.ai cost?

Per-API-call pricing, scaled to volume. Most operators using Captur are doing 50k+ parking validations per month; talk to your account manager for an intro.

Is the city MDS export free?

Yes for most plans. Some large-fleet plans have an add-on charge for direct city API access (where cities pull data programmatically rather than receive manual exports).

Rider experience

Can a rider refuse the helmet selfie?

In strict mode, no — the unlock is blocked. With strict mode off, an ambiguous event still starts the ride. We're considering an opt-out path with a higher per-minute rate, but it's not shipped yet.

What if the rider doesn't have a smartphone camera?

This is a hard requirement — Levy Vision needs the camera. If you have a low-tech rider segment, leave helmet enforcement off for their subaccount.

Will riders see the AI's reasoning?

Riders see the verdict ("helmet required" or "parking issue detected") plus the action (retake, fee, appeal). They do not see the model's confidence band or internal reasoning string — that's operator-only.

Can riders dispute every fee?

Yes — every parking-pose fee has an appeal path with a 48-hour SLA. Sidewalk-event fees can be disputed by adjusting the ride fare via the normal support flow.

Vendor & integration

Should I pick Drover or Luna?

For US fleets, Drover is the default — broader OEM coverage and stronger sales presence. For EU fleets, Luna is often the better choice (closer support, EU data hosting). Either works on Levy Fleets; we don't have a preference beyond your fit.

Can I switch vendors later?

Yes, but you'll need to replace hardware (vendor-specific modules) and re-provision devices in Safety > Vendor devices. Events from the old vendor stay in the events history; new events come from the new vendor.

What OEMs are supported for sidewalk detection?

Depends on the vendor. Common confirmed: Segway Max series, OKAI ES400/ES600, NIU KQi, Bird One. Confirm with the vendor for your specific model before procurement, especially for OKAI ES400 where mid-ride speed-cap can be firmware-dependent.

Can I use my own on-vehicle hardware?

Not today. Phase 3 (own Levy-branded hardware module) is on the roadmap but not committed. If you're considering custom hardware, talk to us — there's a protocol scaffold for it on our side.

City compliance

Which cities accept Levy Vision feeds?

Any city that consumes MDS. That includes most major US cities (Chicago, SF, LA, Atlanta, Austin, Seattle) and many EU cities (Paris, London, Berlin, Amsterdam). For cities with custom schemas, we can usually adapt — contact support.

Can a city require Levy Vision as a permit condition?

Some are starting to. Chicago's 2023 RFP scored on-vehicle sidewalk detection. Paris, SF, and London are heading the same direction. The export covers the documentation cities ask for.

Can I share the MDS export with my insurance broker?

Yes. The export is your data — share it with anyone who needs it. Many operators use it to negotiate insurance premiums with their broker (Christensen Group is a common partner for this).

Operations

Do I need to retrain my fleet techs?

Minimal training. The main new workflow is Safety > Parking reviews for appeal resolution. Plan 30 minutes to walk a tech through the queue + approve/reject workflow. The rest of the dashboard is read-only for techs.

What's the recommended pilot duration?

Three to four weeks. Week 1 enable, week 2 watch events, week 3 tune thresholds, week 4 enable strict mode and / or throttle cut. Most operators are at GA-ready by week 4.

Will Levy Vision slow down my unlock flow?

Helmet adds 2–4 seconds. Parking-pose adds 0 seconds (the photo is already collected). Sidewalk detection adds 0 seconds (it's on-vehicle, async). Most riders don't perceive the helmet delay after the first few rides.

What happens to the events when I disable Levy Vision?

Existing events stay in the database and the dashboard. New events stop being generated. If you re-enable later, you pick up where you left off; no backfill of the gap.