When drivers quit the app, detection recovers
Force-quit recovery within seconds. The driver doesn’t reopen the app — the trip completes.
Automatic recovery
No driver action
Trip always completes
Your behavior change program needs three things scored on every trip — phone use, hard braking, and rapid acceleration. Sampark is the part that makes sure all three are captured, on the devices your drivers actually carry, without a trip getting missed.
Three signals that behavioral research and field data consistently identify as accident precursors. Accuracy on a narrow target beats guesswork across a broad one.
ML on-device identifies active phone use — texting, scrolling, calling without hands-free — and separates it from incidental handling. No network required. A fleet manager can’t coach a driver on something they can’t see. This makes it visible.
Detected from accelerometer and gyroscope, time-aligned with GPS. Threshold calibrated to your fleet before the program goes live — a delivery van on a slope is not a motorcycle at speed. Hard braking responds to coaching faster than any other signal.
Same sensor pipeline as hard braking. Aggressive acceleration after stops predicts hard braking later in the same trip — they’re behaviorally paired. Drivers who score high on both are the highest-priority intervention targets.
Speeding is the next signal we are adding to the pipeline. We are not claiming it until the model is validated. That is the standard we hold for everything on this page.
Most SDKs are tested on flagship devices. Most fleets run on budget Android OEMs — the ones whose process killers break background detection within weeks of deployment. We built Sampark on those devices first.
Its game-mode and task-manager utilities kill background apps within 10 minutes of inactivity. We use a persistent foreground service plus activity-recognition wake-ups to survive both.
Auto-start permission, battery saver lockout, and background restriction stacked three layers deep. Our install flow walks the user through three explicit toggles.
Inherits the same aggressive memory policy as Tier 2, with stricter app-lock defaults. Same fix pattern, with one extra prompt for app-lock unbinding.
Stock Android and iOS — the easy case. Used as the calibration reference; if it fails here, we don’t ship.
Currently calibrating against pilot fleet. Its background-restriction layer sits above the Android one; expected to ship calibrated by Q3 2026.
Every behavioral profile is built from trips. A missed trip is a gap in the coaching conversation. Here’s how we ensure trips complete.
Force-quit recovery within seconds. The driver doesn’t reopen the app — the trip completes.
One integration, three signals, every trip — not partial data, not separate SDKs to combine.
Thresholds calibrated against your pilot data — your devices, your geography, your drivers. Not the number that looked good in the demo.
Stored on-device, uploaded when signal returns. Rural areas, tunnels, low-signal zones — no trip is lost.
One init call, same API surface on iOS and Android. If your team has integrated CMT or Zendrive before, the shape will be familiar. The code samples below are the actual integration — not a simplified version.
Sampark.configure(...) is everything.// iOS · Swift import Sampark let sampark = Sampark.configure( tenant: "acme-insurance", region: .in_south, onTripStart: { trip in analytics.log("trip.start", trip.id) }, onTripEnd: { trip in backend.submit(trip) } ) sampark.setUser(id: currentDriver.id) sampark.start()
// Android · Kotlin import com.sastram.sampark.Sampark val sampark = Sampark.configure( tenant = "acme-insurance", region = Region.IN_SOUTH, onTripStart = { trip -> analytics.log("trip.start", trip.id) }, onTripEnd = { trip -> backend.submit(trip) } ) sampark.setUser(currentDriver.id) sampark.start()
// Backend · trip.completed payload { "version": "2", "event": "trip.completed", "trip": { "id": "trip_01HZ8X7K4M...", "user_id": "drv_aaa111", "start_at": "2026-05-03T08:42:11Z", "end_at": "2026-05-03T09:05:38Z", "distance_km": 12.4, "score": { "overall": 82, "smoothness": 88, "focus": 76 }, "placement": "mounted", "transport": "car" }, "signature": "sha256=..." }
A detection gap is a profile gap. Every SDK has failure modes. Vendors that don’t name them haven’t shipped at scale.
We save trip data locally as it comes in, but at some point the phone running low on storage or the OS clearing old data wins. Trips after day 14 may be partially lost on aggressive devices.
Fix: Lower-frequency local heartbeat & explicit user prompt at day 7.
Idle CPU is 0%. But a 90-minute daily commute with hard maneuvers will use measurable battery — on the order of 3-5% per hour of active driving.
Fix: We show you the per-driver envelope on call, not a marketing average.
The big one. Most SDKs lose detection when the user swipes the app away. Sampark uses geofence wake-ups and activity-recognition transitions to recover within 30-90 seconds of motion.
Fix: Layered recovery, validated across 1,00,000 km of real-world data.
30 minutes with the engineer who’d own your integration. Bring your device list — we’ll tell you exactly what to expect, including where the SDK will degrade.
Book a call →