📷 Identify 📚 Library 🗺 Map 📓 Journal 📄 Docs ℹ About
huggingface-expert · Claude Code Skill

The Botanical Intelligence Pipeline

End-to-end ML inference for plant identification, medical entity extraction, chemical compound detection, and evidence-based Q&A — entirely local, entirely private.

6 ML Models
14k Plant Species
41 Entity Types
100% Local
01

The Pipeline

Five-stage inference chain from photograph to evidence-backed botanical report.

◉ Concept Architecture
📷
Photo Input
Camera / upload
→
344 MB
🌿
Plant ID
PlantNet ViT
image-classification
→
713 MB
⚕
Medical NER
41 entity types
token-classification
→
6.3 GB
📚
BioGPT QA
PubMed evidence
text-generation
→
✔
Results
Species + entities
+ evidence report
▣ BPMN 2.0 Process Diagram
USER INPUT INFERENCE GATE ▶ Start ? Image? Identify Plant PlantNet ViT yes Enter Plant Name manual input no KB Lookup herb_library.py Medical NER 41 entity types BioGPT Evidence PubMedQA check_resources.py RAM + disk + model size gate GREEN YELLOW RED = HALT End
Model task Decision / gate Sequence flow Resource check
02

Models

Six components — five ML models and one safety gate script.

Vision 344 MB

PlantNet ViT

janjibDEV/vit-plantnet300k

Vision Transformer trained on the Pl@ntNet-300K dataset. Identifies approximately 14,000 plant species from photographs with high accuracy.

image-classification Apache-2.0
NER 713 MB

Medical NER

blaze999/Medical-NER

Extracts 41 medical entity types from text: medications, diseases, dosages, procedures, lab values, and more.

token-classification MIT
Chemistry 290 MB

ChemicalDetect

OpenMed/ChemicalDetect

Identifies active chemical compounds and molecular entities in botanical and pharmacological text descriptions.

token-classification Apache-2.0
QA 6.3 GB

BioGPT PubMedQA

microsoft/BioGPT-Large-PubMedQA

Large language model fine-tuned on PubMed literature. Answers yes/no/maybe questions about clinical evidence for herbal compounds.

text-generation MIT
NER 508 MB

Biomedical NER

d4data/biomedical-ner-all

General-purpose biomedical named entity recognition. Extracts diseases, chemicals, genes, species, and cell types from literature.

token-classification Apache-2.0
Safety Script

Resource Gate

scripts/check_resources.py

The load-bearing rule. Every model download and every model load MUST pass through this gate. Returns green, yellow, or red based on available RAM and disk space.

check_resources.py Never bypassed
03

Herb Index

Reference catalogue of key medicinal herbs supported by the pipeline.

🌿

Stinging Nettle

Urtica dioica
Leaves, root, seeds
BPH/prostate support, seasonal allergies, anti-inflammatory, iron-rich nutritive tonic.
quercetin, beta-sitosterol, scopoletin, histamine
🌼

Chamomile

Matricaria chamomilla
Flower heads
Sleep aid, anxiety relief, digestive calming, anti-inflammatory skin wash.
apigenin, bisabolol, chamazulene, matricin
🌻

Dandelion

Taraxacum officinale
Root, leaves, flowers
Liver tonic, gentle diuretic, digestive bitter, blood sugar support.
taraxacin, inulin, luteolin, chicoric acid
🍇

Elderberry

Sambucus nigra
Berries, flowers
Cold and flu remedy, immune support, antiviral activity, fever reduction.
anthocyanins, quercetin, rutin, sambucol
🎵

Valerian

Valeriana officinalis
Root
Insomnia treatment, anxiety reduction, muscle relaxant, nervous system sedative.
valerenic acid, isovaleric acid, GABA modulation
☀

St John's Wort

Hypericum perforatum
Flowering tops
Mild-to-moderate depression, nerve pain, wound healing, antiviral.
hypericin, hyperforin, flavonoids, xanthones
☘

Yarrow

Achillea millefolium
Aerial parts, flowers
Wound staunching, fever management, digestive aid, menstrual regulation.
achilleine, azulene, camphor, tannins
🍃

Plantain

Plantago major
Leaves, seeds
Wound poultice, insect bite relief, cough and bronchial support, drawing agent.
aucubin, allantoin, mucilage, tannins
❤

Hawthorn

Crataegus monogyna
Berries, leaves, flowers
Heart tonic, blood pressure regulation, circulatory support, antioxidant.
vitexin, oligomeric proanthocyanidins, quercetin
🌴

Saw Palmetto

Serenoa repens
Berries
BPH/prostate support, 5-alpha reductase inhibition, urinary flow improvement.
fatty acids, phytosterols, beta-sitosterol, lauric acid
04

Usage Guide

How to invoke the pipeline via Claude Code, CLI scripts, or direct Python.

Via Claude Code

Ask Claude Code in natural language. The skill auto-discovers, gates, downloads, and runs the right model.

natural language 
# Identify from a photo
"Identify this herb from my photo at ~/garden/IMG_4021.jpg"

# Look up medicinal properties
"What are the medical uses of stinging nettle?"

# Run full pipeline on a name
"Run the herb pipeline on Matricaria chamomilla"

# Evidence-based question
"Is there clinical evidence for valerian treating insomnia?"

Herb Pipeline Direct

Call the herb pipeline script directly for batch or automated processing.

bash 
# Full pipeline: identify + NER + evidence
.venv/bin/python3 scripts/herb_pipeline.py \
  --image ~/photos/herb.jpg

# Pipeline from name (skip image classification)
.venv/bin/python3 scripts/herb_pipeline.py \
  --name "Urtica dioica"

# Output as JSON
.venv/bin/python3 scripts/herb_pipeline.py \
  --name "chamomile" --json

Individual Scripts

Each pipeline stage is a standalone script you can invoke independently.

bash 
# Search HuggingFace Hub
.venv/bin/python3 scripts/search.py "medical NER"

# Inspect a model's metadata
.venv/bin/python3 scripts/inspect.py blaze999/Medical-NER

# Gate check before download
.venv/bin/python3 scripts/check_resources.py --model-size 713

# Download a model
.venv/bin/python3 scripts/download.py blaze999/Medical-NER

# Run inference
.venv/bin/python3 scripts/run.py blaze999/Medical-NER \
  --text "Nettle leaf extract 500mg daily"

Installation

Bootstrap the environment and run a smoke test.

bash 
# 1. Clone or enter the project
cd bacon-ai-huggingface

# 2. Run the installer (creates .venv, installs deps)
bash install.sh

# 3. Smoke test: medical NER demo
bash examples/medical_ner_demo.sh

# 4. Verify the venv
.venv/bin/python3 -c "import transformers; print('OK')"
⚠

Resource Gate — The Load-Bearing Rule

Every download and every model load must pass through scripts/check_resources.py first. The gate returns one of three verdicts: GREEN (proceed), YELLOW (warn, allow with caution), or RED (halt, insufficient resources). This rule is never bypassed without explicit user authorisation.