About

About this system and how to use it safely

An intelligent system for medicinal plant identification and symptom-based herbal treatment recommendation, built as a final-year Computer Science major project.

What this system does — and does not — do

It does

  • Identify a plant from an uploaded photo through a model-ready pipeline
  • Match everyday symptom descriptions to recorded traditional uses
  • Explain every score with the keywords that produced it
  • Present botany, properties, remedies and precautions together
  • Save analyses, collect feedback and produce a printable report

It does not

  • Diagnose any disease or condition
  • Prescribe treatment, dosage or duration
  • Replace a doctor, pharmacist or registered practitioner
  • Guarantee that a plant identified from a photo is safe to consume
  • Detect herb–drug interactions specific to your medication

Technology

Front end

React 19, TypeScript, Tailwind CSS and accessible shadcn-based components.

API & ML

FastAPI serving a MobileNetV2 transfer-learning classifier (224×224 input) and a TF-IDF + cosine similarity recommender.

Data & storage

Curated plant, symptom and remedy datasets served by the API, with history and feedback in the cloud database.

Known limitations of this build

  • Results are produced only by the analysis service. If the trained model or the fitted symptom matcher is not installed, the affected page says so instead of showing a result.
  • No accuracy figure is claimed anywhere in the interface; the numbers shown are the model's own confidence and similarity values for your input.
  • Grad-CAM heatmaps are computed from the trained model itself, and nothing is drawn when the model does not return one.
  • The knowledge base only contains what the curated plant dataset contains, so many symptoms will legitimately return no match.
  • History and feedback are stored in your account, so they are only visible when you are signed in.

Safety questions