Skip to content
INSPECTION SOFTWARE

EquipCert AI

Web and mobile inspection evidence with offline capture and explicit certification boundaries.

EquipCert AI public website screenshot
CURRENT SCOPE

Deployed inspection app · operational probes verified

WHO IT'S FOR

Independent build

My web/mobile equipment-inspection product. Operational probes and tests have dated evidence; regulatory certification is not claimed.

THE ENGINEERING

An equipment-inspection application with Next.js, Flutter, Supabase records, offline evidence capture, signatures, optional AI assistance and PDF reporting. October 14 verification recorded 104 web unit tests, seven production probes, three dependency checks and zero configuration-boundary violations across 157 files. The database keepalive now requires an actual database query. The 429 native-client tests and mobile build results remain dated October 1 evidence. Authenticated inspection walkthroughs, persisted signature-consent evidence, commercial payments, an off-machine signing-key backup and a completed restore drill remain open; inspection evidence is not regulatory certification.

Next.jsFlutterSupabase

BEFORE → AFTER

Real problems from this build, what I changed and the verified result.

  1. CASE 01
    Before

    The audit table existed but nothing wrote to it, application users could insert rows, and its append-only guard would have crashed GDPR erasure the moment it filled up.

    After

    Database suite went from 8 failing to 28 of 28 passing, erasure included.

    How I fixed it

    Moved audit writing into database triggers on the core tables, blocked application writes, and defined the one permitted erasure mutation.

  2. CASE 02
    Before

    The live database went offline: the free hosting tier paused it after a week without activity, even though a keep-alive job reported success twice a week.

    After

    October 14 verification passed seven production probes, three dependency checks and 104 web unit tests. Authenticated inspection and a complete restore drill remain separate gates.

    How I fixed it

    Found that the keep-alive called a health endpoint that never touched the database. It now runs a real query every day and reports success only when a row comes back.

  3. CASE 03
    Before

    On the serverless host, the subscription webhook would have rejected every genuine payment event: the platform parsed the body first, but the provider signs the exact original bytes.

    After

    With provider-signed test payloads, the old handler fails the platform-replay and raw-stream cases and the new one passes all of them; serverless behavior was checked against the platform runtime's source.

    How I fixed it

    Made the handler read the exact bytes on both deployment targets, and stopped logging event contents that can include customer details.

Evidence reviewed: Source review October 14, 2026; verification dates and open gates stated above