Turn the inspection footage you already have into decisions

existing footage in — detections, measurements, counts, 3D, report & work-order export out

ScanHawk's inspection-intelligence platform ingests the media you already captured — drone, helicopter, fixed camera, field phone, thermal, or aerial — and turns it into severity-ranked findings, measurements, asset counts, a 3D model, and an AI inspector you can ask anything. No new capture. No fixed defect list.

Ask for any component or defect in plain English — antenna corrosion, soiled solar modules, insulator flashover, blade erosion, vegetation encroachment, a stamped serial — and the platform finds it, measures it, and exports it straight into your report, map layer, and work-order system. The live sections below let you drive it yourself.

§01

#What you get

Send the footage you already have. Get back decisions — not a video to re-watch.

Severity-ranked findings

Every defect and component found, ranked critical to low, with a confidence score and the exact frame it was seen.

Measurements & counts

Corrosion area, erosion length, soiling coverage, asset counts — the numbers a crew or adjuster prices against.

3D model

A navigable 3D reconstruction of the asset from the same footage, for context and change-over-time.

Branded report

A clean, severity-ranked report you can forward to a client, board, or regulator — your logo, optional white-label.

AI inspector

Ask the footage anything in plain English — count the antennas, measure the corrosion, is this safe to climb — and get a grounded answer.

GIS & work-order export

Findings drop into your map layer and work-order system with the location anchor intact — finding to dispatch, no re-keying.

§00

#Enterprise Visual Data Demo

Upload video and make it LLM-ready

Public demo preflight for uploaded inspection video. The browser checks size, type, duration, and resolution, then stages the async path that turns raw media into an inspection memory the AI inspector can answer from quickly.

public web demo

250 MB / 5m00s

MP4, MOV, or WebM - one video

gated pilot

1024 MB / 20m00s

multipart upload + async notification

Inspection videoSource typeInspection program

selected video

utility-corridor-sample.mp4

preflight passed

size
184.0 MB

duration
3m59s

resolution
1920x1080

source
Drone / aerial video

program: Grid asset and vegetation review

Public demo limit passed: under 250 MB and under 5 minutes.
Production path uploads direct to S3, then queues async extraction.
Expected LLM-ready output: inspection-memory.json, keyframes, report rows, and work-order export.

processing stage

Raw video to inspection memory

queued

Preflight

Validate type, size, duration, resolution, checksum, and run metadata.

Store

Direct-to-S3 upload in production; the app queues processing with an object key.

Normalize

Create preview MP4, thumbnails, and a downscaled processing proxy when needed.

Sample

Take 1-2 fps base frames plus adaptive bursts around motion, scene changes, and risk.

Extract

Run detection, OCR, masks, tracking, optional 3D, and geospatial attachment.

Memory

Build frame, track, finding, severity, unknowns, map, and work-order indexes.

Ask

The AI inspector answers questions about the whole inspection with evidence timestamps and UI actions.

Samples

pending

Findings

pending

Map pins

pending

Work orders

pending

LLM-ready artifact

inspection-memory.json

frame ledger, track index, severity-ranked findings, evidence timestamps, unknowns, map context, work-order rows, and compact retrieval text for the AI inspector.

id	asset	finding	sev
F-1042-01	STR-1042	Suspension insulator chip	critical
F-1044-02	STR-1044	Crossarm corrosion	high
F-1045-03	SPAN 1044-1045	Vegetation encroachment	medium

Handoff

LLM-ready output

Public demo uses local preflight. Production uses presigned S3 upload, queued extraction, then the AI inspector answers from the saved memory instead of reprocessing raw video on each question.

sample intake app export

§02

#Pipeline

ingest

your existing footage

→

inspect

detect · measure · read

→

track

count across frames

→

model

3D reconstruction

→

deliver

report · GIS · work orders

§03

#Runs

Sample inspections processed end-to-end across verticals — telecom, solar, utility, renewables, agriculture, industrial. Click a row to open it live.

id	label	vertical	source	duration	res	notes
demo-001	tower_cu_01	telecom	youtube/TFCwRzQ5lVI	3m19s	1280×720	close-orbit DJI Mavic Pro, single tower, antenna cluster + RRU
demo-002	tower_wide_02	telecom	youtube	1m50s	1280×720	wide-shot monopole, sky + rural field background
demo-003	tower_topdown_03	telecom	youtube	1m15s	1280×720	top-down view of antenna cluster — unique angle, radial panel layout
demo-004	solar_pv_01	solar	youtube	1m47s	1280×720	PV farm aerial pass — soiling, hotspot, cracked-cell flags
demo-005	trans_line_01	utility	youtube	1m03s	1280×720	HV transmission tower aerial — insulator string + structural members
demo-006	wind_blade_01	renewables	youtube/AV1isLII4TA	0m27s	720×1280	DJI Matrice 4T 112x zoom — leading-edge erosion + lightning strike
demo-007	crop_yield_07	agriculture	youtube	2m25s	1280×720	precision agriculture — tractor + forage harvester + crop rows / NDVI baseline
demo-008	livestock_08	agriculture	youtube	1m09s	1280×720	livestock counting + herd cluster detection over pasture
demo-009	cylinder_ocr_01	industrial	field photo	still	still image	stamped propane-cylinder serial — AI-vision OCR with per-character confidence + ambiguous-glyph flags

§04

#Run / demo-005

label trans_line_01vertical utilityduration 1m03sres 1280×720

This is a sample run on demo media. Run the same AI stack on your inspection media and get a severity-ranked report, map pins, and work-order export.Analyze your media ->

task mode

ANALYZING · Detection

t=0.0s · 0 detections

inspection assistant · ScanHawk AI

Ask in plain language what you want to see, count, or measure. The assistant routes your request to the right capability live — detect & count, measure an area, read a tag — and can turn the result into a branded report.

loading whole-video inspection memory...

custom prompt · runs on the current video frame

prompt cookbook · click to use

backend: runs on your footage in our secure cloud · sub-5s once warm

worker.detect(img, [components+damage])

prompt: "Locate all the instances that match the following description: [components + damage descriptions]."

Multi-class open-vocabulary detection. Returns bounding boxes for components and damage flags — find anything you can describe in plain English.

3D reconstruction

INITIALIZING

24 keyframes·feed-forward · 6.2s·scale norm @ p95·~53k pts

detections.json

/detections/demo-005.json

Per-frame bounding boxes with track_id continuity, label, severity, confidence. Coordinates normalized [0,1].

cloud.ply

/3d/demo-005.ply

Sparse point cloud, ASCII PLY format. Centered + scale-normalized. ~80k points.

poses.json

/3d/demo-005.poses.json

Per-keyframe camera position and orientation data.

§06

#Why ScanHawk

No new capture

Runs on footage you already paid to collect — drone, helicopter, fixed camera, field phone, thermal. Nothing to re-fly.

Ask for anything, in plain English

Not a fixed list of defect classes. Describe what you care about — flashover on an insulator, a stamped serial, leading-edge erosion — and it finds it.

Measures, doesn't just flag

Defects come back as area and length, not just a box — the number a vendor quotes against and an adjuster needs.

Plugs into your system of record

GIS / Esri and EAM / work-order export with the location anchor intact, so a finding becomes a dispatched crew.

Reads the hard stuff

Stamped and embossed serials, nameplates, and asset tags that ordinary OCR returns nothing for — with per-character confidence.

Built for enterprise

Capture-agnostic and hardware-neutral, with a security and data-handling posture built for utility and infrastructure programs.

Run the AI stack on your footage →