Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/elzackarias/Hackaton3B-Reto1/llms.txt

Use this file to discover all available pages before exploring further.

Anaquel Inteligente 3B is a real-time shelf inventory monitoring system built for Tiendas 3B convenience stores. A YOLOv8-seg model watches a camera feed trained on 7 specific SKUs, detects when customers pick up or return products, and immediately updates inventory counters — all without QR codes, RFID tags, or manual intervention. Results stream live to a React dashboard via WebSocket, including depletion predictions, a slot activity heatmap, and intelligent narrative alerts in Spanish.

Quickstart

Set up the backend and frontend, run the server, and see live inventory updates in under 10 minutes.

Architecture

Understand the full data pipeline from camera frame to dashboard event.

API Reference

Explore all REST endpoints and WebSocket events exposed by the FastAPI backend.

Detection Engine

Learn how YOLOv8-seg detects products and generates inventory change events.

How It Works

1

Camera captures frames

A fixed 5 MP camera (USB or RTSP) streams frames at up to 5 FPS. The CameraCapture module handles reconnection automatically.
2

YOLOv8-seg detects products

The DetectionEngine runs inference on every frame and counts visible instances of each of the 7 trained SKUs. A 3-frame consistency check and per-slot cooldown prevent false positives.
3

InventoryEngine updates stock

When a removal or return is confirmed, InventoryEngine.process_event() atomically updates stock counters, triggers alerts when stock falls below the configurable threshold (default 20%), and fires observer callbacks.
4

Intelligence engines enrich data

PredictionEngine forecasts time-to-stockout using exponential smoothing. HeatmapEngine aggregates slot activity. NarrativeEngine generates human-readable status messages in Spanish.
5

Dashboard receives live updates

FastAPI + Socket.IO broadcasts inventory_update, prediction_update, heatmap_update, narrative, and video_frame events. The React dashboard renders them in real time.

Key Features

Computer Vision

YOLOv8-seg trained on 7 real Tiendas 3B products. Detects product counts per slot with bounding boxes and confidence scores.

Anti-False-Positive

3-frame consistency check plus a per-slot cooldown timer prevent double-decrements and flickering detections.

Depletion Prediction

Exponential smoothing (α=0.3) over removal timestamps predicts minutes until stockout per SKU.

Activity Heatmap

Slot interaction counts within a configurable time window, returned as a structured JSON map.

Narrative Alerts

Auto-generated Spanish messages for removals, returns, threshold breaches, and predictive warnings.

Smart Restock

Priority-scored restock queue combines units missing, sales velocity (iventa), and fill rate.

Build docs developers (and LLMs) love