All Work
AI SystemsAI Platform2026

LiquorScan — High-Precision Liquor Fill Level Detection Powered by Hybrid Semantic Segmentation

  • PyTorch
  • YOLO
  • FastAPI
  • Computer Vision
  • Streamlit
  • OpenCV
  • AWS
  • Python

The challenge

Manual liquor inventory checks are slow and prone to error. Varied bottle shapes, labels, closures, and reflections heavily distorted single-view computer vision fill level segmentations.

What we built

A PyTorch hybrid semantic segmentation core (ResNet-101, FCN, PSPNet), a YOLO object detection pipeline, a multi-perspective mask quality scoring algorithm, and a stateless FastAPI app on AWS.

Systems delivered

  • Computer Vision Core
  • Hybrid Semantic Segmentation Pipeline
  • Multi-Perspective Mask Quality System
  • FastAPI Backend Service
  • Streamlit Interface

Outcome

Automated liquor inventory checks in ~9–10 seconds with high precision across 16 segmented classes, eliminating subjective manual visual estimations.

Want this as a PDF?

Download the full case study to share with your team.

Have a similar product challenge? Let's talk about the right technical path.

Discuss a Similar Build