VisionPro

01The Problem & Business Context

In retail and quick-service operations, reconciling food and goods handed across the counter with registered transactions in the cashier system is an ongoing operational challenge. Discrepancies often go unnoticed until late inventory audits, making it difficult to distinguish cashier input errors from unrecorded goods.

Working directly with a business owner based in Morocco, we sought to automate the daily verification process: capturing physical items passed across the service area and comparing those events with time-stamped register receipts to generate an end-of-day summary of total goods served.

02Product Workflow

1. Counter Video Capture: RTSP network camera streams capture the handover zone at the counter.
2. Local Event Detection: Frame streams are ingested on-premises by an edge mini PC running Frigate, triggering object presence events.
3. Event Publishing via MQTT: Detection events are published over a lightweight MQTT message broker.
4. Data Fetching & Cashier Reconciliation: Core ingestion scripts correlate event timestamps with point-of-sale receipt logs, storing matched records in SQLite.
5. Daily Summary: Provides managers with an end-of-day overview of total items served versus total items billed.

03My Contribution in Collaboration

I worked directly with the business partner from the initial problem breakdown:

• Solution Brainstorming & Architecture: Defined the system boundaries, avoiding expensive cloud AI subscriptions in favor of an on-premises edge computing approach.
• Technical Setup: Configured the RTSP camera streaming pipeline and local network topology for edge hardware.
• Core Data-Fetching Functionality: Authored the core data ingestion logic that listens to incoming event streams, queries local SQLite databases, and prepares reconciliation sets.

04Technical Direction

We selected an on-premises mini PC rather than uploading continuous video to the cloud. This avoids high bandwidth costs, reduces external latency, and preserves customer privacy by keeping raw camera streams strictly inside the physical location network.

05Current Status & Scope Boundaries

VisionPro is an active work in progress. In keeping with honest portfolio standards:

• We do not claim automated theft conviction or facial recognition capabilities.
• The system is an auditing tool designed to assist managers in comparing totals, not an autonomous enforcement agent.
• The pipeline architecture is currently in local staging; full commercial multi-store deployment remains in development.