N.
All work

FungiSenseMushroom Classification System

06 / 07
06
Year
2023
Role
Machine Learning Engineer
Duration
1 month
Client
Vennisys

Source

Overview

Architected a multi-output neural network that simultaneously predicts five mushroom characteristics - species, family, edibility, habitat and season - from physical features. Built with a safety-first design including risk assessment and lookalike warnings.

Challenge

Traditional mushroom identification relies heavily on expert knowledge and visual inspection, creating barriers for educational mycology research, biodiversity cataloging and field-guide applications.

Solution

Engineered a multi-output neural network achieving 98–100% accuracy on safety-critical edibility classification. Implemented weighted loss functions prioritizing safety predictions and built a FastAPI service exposing five specialized endpoints.

Results

  • 01

    98–100% accuracy on edibility classification

  • 02

    40–60ms inference latency per prediction

  • 03

    2.45MB optimized model size for edge deployment

  • 04

    173 species classified across 23 families

  • 05

    Batch processing supporting 100+ predictions per request

Technology

PythonTensorFlow 2.16KerasScikit-learnFastAPIPydanticNumPyPandasDocker