A field inspection app that keeps working when the signal doesn't
A native inspection app that keeps working with zero signal in the field, syncing photos and checklists the moment a connection returns, with push alerts routed through each platform's store-approved channels.
The shape of the work
Home & Field Services
10 weeks
Fixed price
Client name withheld under NDA. Engagement details are shown to the extent our agreement permits.
Offline-First Storage Engine
Persists photos, audio notes and complex checklists locally, with no cell connection needed.
WatermelonDB over SQLite holds the whole inspection locally (photos, audio notes, nested checklists), so the app behaves identically with or without signal. There's no read-only mode to degrade into. Photos are compressed and written to disk on capture, never held in memory, which is what lets an inspector shoot four hundred frames in a day without the OS killing the app.
- Full local model: photos, audio and nested checklists
- Photos compressed to disk on capture, never held in memory
- Identical behavior on and off signal
The fault sheet filled in at the enclosure: six readings against their limits with three over, three photographs and a voice note on the handset, the hazard flag set, and all of it written locally.
Automatic Reconnect Sync
Signal returns and the queue empties itself: seven inspections and twenty-two photographs acknowledged by the server 6.5 seconds later, zero records lost, and one verdict held back for a supervisor.
Background sync with automatic conflict resolution, clearing in under 8 seconds once signal is back.
Sync is a background task that resumes where it stopped, so a truck driving in and out of coverage makes steady progress instead of thrashing. Conflicts resolve per field, with last-write-wins on scalars and union on collections, and anything genuinely ambiguous is queued for a supervisor. Nothing gets silently picked. A full day's backlog clears in under eight seconds on a normal connection.
- Resumable background sync that picks up where it dropped
- Per-field resolution: last-write on scalars, union on lists
- Ambiguous conflicts queued for a supervisor, never guessed
Urgent Safety Push Alerts
APNs and FCM integration for instant routing of flagged utility hazards to supervisors.
A flagged hazard doesn't wait for the sync queue. It takes a dedicated priority path straight to APNs and FCM the moment any connectivity appears, carrying the location and a photo thumbnail in the payload so a supervisor can triage from the lock screen. Delivery is acknowledged back to the device, and an unacknowledged alert re-sends. Nothing is assumed received.
- Priority path that jumps the ordinary sync queue
- Location and thumbnail carried in the notification payload
- Unacknowledged alerts re-send until delivery is confirmed
The hazard's priority path: the alert as the supervisor's lock screen shows it, with location and photo thumbnail, sent ahead of the queue, and an APNs send with no delivery receipt re-sent instead of assumed delivered.
What we were brought in to do
Inspectors work substations and pipelines with no cell coverage for hours at a time. We built a native iOS and Android app that captures full inspections offline and reconciles everything the moment a signal returns.
A utility's inspection contractor covering substations and pipeline runs where crews are out of coverage for four to six hours at a stretch. The previous app had been written for an office network and taken into the field unchanged. The engagement began after an audit found inspection records with photographs missing and no way to tell whether they had ever existed.
Mobile Engineering
Where the old way broke
The previous app assumed connectivity and silently dropped photos and checklist entries in dead zones, so inspectors redid work back at the office and supervisors couldn't trust the record.
The old app buffered in memory and posted on save. In a dead zone the post failed, the buffer was discarded on the next screen, and the interface reported success, so an inspector genuinely believed the work was recorded. Supervisors had learned not to trust the record, which meant crews were re-photographing sites they had already documented.
We built an offline-first data layer that persists every photo, note and checklist entry locally, then syncs and resolves conflicts automatically once the device reconnects. Push notifications run through APNs and FCM for urgent flagged issues.
What we built together
- 01
Designed local-first storage so a full inspection survives a dead zone entirely
Local-first was scoped against the routes with the worst coverage, because the worst route is the one that decides the design.
- 02
Built background sync with conflict resolution for entries edited on two devices
Sync resumes where it stopped, and conflicts resolve per field (last write on scalars, union on collections), with anything genuinely ambiguous queued for a supervisor.
- 03
Wired push notifications through APNs/FCM for flagged safety issues
A flagged hazard takes a priority path that jumps the sync queue entirely, carrying the location and a thumbnail so a supervisor can triage from the lock screen.
- 04
Prepared signing, store metadata, and privacy declarations for a clean first review
Signing, store metadata and privacy declarations were prepared alongside the build, which is why both reviews passed on first submission.
Phase by phase
Phase 1: Field Audit & Dead Zone Specs
Zero-Signal Offline Architecture
Accompanied utility inspectors into remote substations to map offline photo capture, audio note recording, and checklist persistence requirements.
- Offline Architecture Spec
- Dead Zone Usability Matrix
- Data Schema
Phase 2: Local Persistence Layer
SQLite & Sync Engine Engineering
Engineered local SQLite and WatermelonDB storage capable of holding hundreds of high-res inspection photos offline with automatic background sync.
- Local Sync Engine
- Conflict Resolution Logic
- Storage Encryption Layer
Phase 3: Push Alerts & Native Build
Safety Alerts & APNs/FCM Integration
Integrated APNs and FCM push messaging to route urgent safety flags and hazardous site warnings directly to supervisors.
- Push Notification Pipeline
- Native iOS & Android Builds
- Safety Alert Rules
Phase 4: Store Submission & Deployment
App Store Review & Fleet Rollout
Prepared store metadata, privacy declarations and code signing, and passed Apple App Store and Google Play Store review on the first attempt.
- App Store & Play Store Clearances
- MDM Fleet Deployment
- Inspector Onboarding
The same check answered twice, once offline in the field and once in the office, resolved per field: the reading takes the last write, photographs and notes keep both, and the disagreeing verdict waits for a supervisor.
Operational results after launch
0
Inspections lost to dead zones
< 8s
Sync time on reconnect
Passed 1st try
Store review
Inspections lost is a count from the reconciliation ledger over the first six months, not a survey. Sync time is measured from the device regaining signal to a full day's backlog being acknowledged by the server, at the ninetieth percentile. Store review is the first submission on each platform.
Client name withheld under NDA. Figures are approximate, drawn from the engagement’s own reporting.
About our collaboration
A cross-functional team of 4 worked on a fixed price basis over 10 weeks, covering Native app build, Offline-first sync, Store release. We shipped in two-week increments, each one releasable and reviewed live before it merged. Decisions were written down as they were made, so the reasoning outlived the people who made it.
The build was tested in the field: two inspectors carried builds on real routes from week three, including the routes with the worst coverage. Store submission was prepared alongside the work, with signing, metadata and privacy declarations in place before the first release candidate, which is why review passed the first time.
What we'd carry into the next one
Zero inspection loss in dead zones ended the rework and the supervisors' skepticism.
The rework disappeared along with the uncertainty. Crews hadn't been re-photographing sites because records were lost, but because nobody could tell which ones were.
Background conflict resolution synced inspection photos and checklists in under 8 seconds upon reconnect.
Resumable sync mattered more than fast sync. A truck moving in and out of coverage makes progress on every fragment, where a restart-on-drop design would start over each time.
Compliant store release pipelines passed Apple App Store and Google Play Store reviews on first submission.
Preparing the store submission during the build turned two weeks of unpredictable review cycles into a single pass on each platform.
One round, through a dead zone
Five and a half hours without signal, seven inspections, and nothing that depends on the network to exist.
Step through the day on one handset: the signal drops, every check and photograph is written locally, the signal comes back and the hazard leaves first, then the queue drains to the server. Pick a moment, or use the arrow keys once a tab is focused.
7 inspections · 22 photographs · 5 h 38 min without signal
Row-by-row timings are illustrative; the 6.5 s drain is the recorded run
- Line 14 · MLV-23 valve stationINS-4466 · 8:06 AM · 3 photosPlanned
- Line 14 · MLV-26 valve stationINS-4467 · 8:51 AM · 2 photosPlanned
- Pike Road regulator stationINS-4468 · 9:40 AM · 3 photosPlanned
- Harlan Creek Sub · Transformer T2INS-4469 · 10:31 AM · 5 photosPlanned
- Harlan Creek Sub · Switchgear bay 2INS-4470 · 11:28 AM · 3 photosPlanned
- Line 14 · CP-31 test postINS-4471 · 12:15 PM · 2 photosPlanned
- Dry Fork pump stationINS-4472 · 12:58 PM · 4 photosPlanned
- 7:48 AMLast bar of signal leaving the depot road
- 7:48 AM7 inspections planned on the round, all cached
The handset is the record. The network only copies it upward.
React Native on iOS and Android, WatermelonDB over SQLite on the device, a Node.js service behind it, and APNs and FCM for alerts. The old app buffered in memory and posted on save; here nothing shows as saved until it's on disk.
- 01 · CapturePhoto, note, checklist entryPhotos are compressed and written to disk on capture, never held in memory, so a long day of frames doesn't get the app killed by the OS.
- 02 · StateWatermelonDB over SQLiteThe whole inspection lives on the handset (photos, audio notes, nested checklists) behind an encrypted storage layer. The app behaves the same with or without signal.
- 03 · QueueResumable background syncResumes rather than restarts, so a truck moving in and out of coverage makes progress. A full day's backlog clears in < 8 s (p90) on reconnect.
- 04 · EnginePer-field reconcile, Node.jsLast write wins on scalars, union on collections. Anything genuinely ambiguous is queued for a supervisor, never silently picked.
- 05 · EdgeAPNs and FCM priority pathA flagged hazard skips the sync queue with location and thumbnail in the payload. Delivery is acknowledged back; no receipt means it re-sends.
A record supervisors can trust
No silent losses, no guessed merges & no alerts lost in transit
Saved means on disk
The previous app buffered in memory, lost the buffer on the next screen, and still reported success. Here every photo, note and checklist entry is persisted on the handset the moment it's captured. Inspections lost to dead zones, counted from the reconciliation ledger over the first six months:0.
Conflicts are resolved, or asked about
When the same entry is edited on two devices, scalars take the last write and collections keep both sides. Anything genuinely ambiguous is queued for a supervisor, so a record never quietly changes its verdict.
A hazard alert is confirmed, never assumed
A flagged safety issue skips the sync queue and goes to APNs and FCM the moment any connectivity appears, with location and a thumbnail. Delivery is acknowledged back to the device; an unacknowledged alert re-sends.
Are your crews re-doing inspections because nobody can tell which records made it back? Scope your build in 3 minutes.
Scope your buildNearby engagements
Web PlatformsA dispatch platform that keeps crews moving
A scheduling and dispatch platform with a technician mobile app, instant quoting, and routing that fits more jobs into every day.
Home & Field Services · 16 weeks
Web PlatformsMatching by who covers the address, not by who is nearest to it
A two-sided marketplace for lawn care where providers draw the area they actually serve, so a job is offered to the people who'll genuinely travel to it.
Landscaping & Lawn Care · 18 weeks
AI & AutomationNever lose a job to a missed call again
A missed-call recovery agent that texts back instantly, captures the job details, books the slot, and keeps a busy calendar full.
Plumbing Services · 6 weeks
Let's talk
Running a large platform, shaping a first MVP, or getting a product ready for a funding round? Tell us where you are. We'll shape the process around it, and stay with you after launch.
