V21.1 • iPhone Field Edition + mobile navigation fix
Pilot-confirmed workflow • no automatic Eberl submissionPlan multiple inspections, geocode addresses, pull property weather, estimate road travel, score Go/No-Go readiness, and launch each job into the field workflow.
V11 uses public map/weather services when available: Photon for address geocoding, OSRM for road travel, and Open-Meteo for weather. Results are advisory and may be unavailable or rate-limited.
Combines job weather, road-travel estimates, appointment timing, and projected pay into a planning score. It does not replace FAA airspace/TFR checks or pilot judgment.
The ZIP now includes openai_server.py, .env.example, and AI_CONTRACT.md. Start the backend, then point this field to http://127.0.0.1:8787/analyze.
V13 turns the inspection requirements into a live shot list. Complete each shot deliberately, then use AI/video as backup coverage.
Import a drone video. V12 samples frames locally, ranks frame quality, rejects near-duplicates, and turns the best frames into inspection candidates for AI classification. Use extracted frames as a backup or supplement unless Eberl specifically accepts video-derived stills for the assignment.
After individual image analysis, V7 summarizes the whole photo set: confirmed category counts, detected roof components, possible observations, QA warnings, and missing capture requirements.
Prioritizes what needs your attention instead of making you inspect every AI result equally.
Generates a concise pre-departure / pre-upload summary from the current job and photo set.
V15 reads GPS/heading metadata from JPEGs when available and can also import a DJI/flight telemetry CSV. Real coordinates replace sequence-based sector guesses whenever possible.
V16 draws the flight path and geotagged captures directly in the browser. It can also bind photos to the nearest telemetry point by timestamp.
V18 lets you define the actual roof/building footprint using geotagged capture geometry or manual points. It then checks whether captures surround the footprint and whether heading data points inward toward it.
V17 builds a real 8-sector coverage model around the estimated property/building center using geotagged photo positions. This replaces sequence-only sector guesses whenever GPS/telemetry is available.
V14 analyzes capture sequence and confirmed photo categories to estimate which roof/property sectors are well covered and where angle diversity is weak.
V9 converts the current inspection state into a rework-risk score and tells you the next highest-value photo/check to complete before leaving.
Save the job state as a portable JSON snapshot so you can replay a test inspection, move between Mac/iPhone, or restore a job without relying only on browser storage.
V19 runs the current inspection through the major QA layers in one guided sequence and produces a single readiness score, prioritized reshoot plan, and submission recommendation.
V20 records what the app predicted versus what actually happened after submission. This lets you tune thresholds using real inspection outcomes instead of assumptions.