{
  "nbformat": 4,
  "nbformat_minor": 5,
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "name": "python",
      "version": "3.12.3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "id": "fe716932",
      "metadata": {},
      "source": [
        "# USDA Phytochemical Database — Quickstart\n",
        "\n",
        "This notebook demonstrates 5 practical use cases with the **free 400-row sample**.\n",
        "\n",
        "**Full dataset (104,388 rows):** [ethno-api.com](https://ethno-api.com)  \n",
        "**GitHub:** [wirthal1990-tech/USDA-Phytochemical-Database-JSON](https://github.com/wirthal1990-tech/USDA-Phytochemical-Database-JSON)\n",
        "\n",
        "---"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "cfb77598",
      "metadata": {},
      "outputs": [],
      "source": [
        "# Cell 2 — Load the 400-row sample\n",
        "import pandas as pd\n",
        "\n",
        "url = \"https://raw.githubusercontent.com/wirthal1990-tech/USDA-Phytochemical-Database-JSON/main/ethno_sample_400.json\"\n",
        "df = pd.read_json(url)\n",
        "print(f\"Records: {df.shape[0]}\")\n",
        "print(f\"Unique compounds: {df['chemical'].nunique()}\")\n",
        "print(f\"Unique species: {df['plant_species'].nunique()}\")\n",
        "print(f\"\\nSchema:\")\n",
        "print(df.dtypes)\n",
        "df.head()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "28718bea",
      "metadata": {},
      "outputs": [],
      "source": [
        "# Cell 3 — Top 10 compounds by PubMed mentions (with bar chart)\n",
        "import matplotlib\n",
        "matplotlib.rcParams['figure.figsize'] = (10, 5)\n",
        "\n",
        "top10 = (\n",
        "    df.groupby('chemical')['pubmed_mentions_2026']\n",
        "    .first()\n",
        "    .nlargest(10)\n",
        "    .sort_values(ascending=True)\n",
        ")\n",
        "\n",
        "top10.plot.barh(color='#2563eb')\n",
        "import matplotlib.pyplot as plt\n",
        "plt.xlabel('PubMed Mentions (March 2026)')\n",
        "plt.title('Top 10 Most-Studied Phytochemicals')\n",
        "plt.tight_layout()\n",
        "plt.show()\n",
        "\n",
        "print(top10.to_frame().to_string())"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "a8e4bde5",
      "metadata": {},
      "outputs": [],
      "source": [
        "# Cell 4 — DuckDB: Anti-inflammatory compounds ranked by clinical evidence\n",
        "# pip install duckdb  (once)\n",
        "import duckdb\n",
        "\n",
        "# Query directly from the pandas DataFrame loaded in Cell 2\n",
        "result = duckdb.sql(\"\"\"\n",
        "    SELECT\n",
        "        chemical,\n",
        "        MAX(pubmed_mentions_2026)  AS pubmed_score,\n",
        "        COUNT(DISTINCT plant_species) AS species_count\n",
        "    FROM df\n",
        "    WHERE application ILIKE '%anti%inflam%'\n",
        "       OR application ILIKE '%antiinflam%'\n",
        "    GROUP BY chemical\n",
        "    ORDER BY pubmed_score DESC\n",
        "    LIMIT 15\n",
        "\"\"\")\n",
        "print(\"Top anti-inflammatory compounds by PubMed evidence:\")\n",
        "result.show()"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "b5a633ab",
      "metadata": {},
      "outputs": [],
      "source": [
        "# Cell 5 — RAG pipeline use case: build a deterministic compound context block\n",
        "# This is the pattern used to ground LLMs with plant chemistry data.\n",
        "\n",
        "query_compound = \"QUERCETIN\"\n",
        "\n",
        "# Filter all records for this compound\n",
        "compound_records = df[df[\"chemical\"] == query_compound].copy()\n",
        "compound_records = compound_records.sort_values(\"pubmed_mentions_2026\", ascending=False)\n",
        "\n",
        "print(f\"=== RAG CONTEXT BLOCK FOR: {query_compound} ===\\n\")\n",
        "for _, row in compound_records.head(5).iterrows():\n",
        "    block = (\n",
        "        f\"Compound: {row['chemical']}\\n\"\n",
        "        f\"Plant: {row['plant_species']}\\n\"\n",
        "        f\"Application: {row['application']}\\n\"\n",
        "        f\"Dosage: {row['dosage']}\\n\"\n",
        "        f\"PubMed mentions (2026): {row['pubmed_mentions_2026']}\\n\"\n",
        "        f\"---\"\n",
        "    )\n",
        "    print(block)\n",
        "\n",
        "print(f\"\\nTotal records for {query_compound}: {len(compound_records)}\")\n",
        "print(\"These structured blocks eliminate LLM hallucinations about botanical dosages.\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "id": "d8220844",
      "metadata": {},
      "outputs": [],
      "source": [
        "# Cell 6 — Parquet vs JSON: memory and size comparison\n",
        "import io\n",
        "\n",
        "json_bytes   = df.to_json(orient=\"records\").encode(\"utf-8\")\n",
        "parquet_buf  = io.BytesIO()\n",
        "df.to_parquet(parquet_buf, engine=\"pyarrow\", compression=\"snappy\")\n",
        "parquet_size = parquet_buf.tell()\n",
        "json_size    = len(json_bytes)\n",
        "\n",
        "print(f\"{'Format':<10} {'Size':>12}  {'Ratio':>8}\")\n",
        "print(f\"{'JSON':<10} {json_size:>12,} B  {'100.0%':>8}\")\n",
        "print(f\"{'Parquet':<10} {parquet_size:>12,} B  {parquet_size/json_size:>8.1%}\")\n",
        "print()\n",
        "print(f\"Parquet is {json_size/parquet_size:.1f}× smaller than JSON for this sample.\")\n",
        "print()\n",
        "print(\"Full 104,388-row dataset:\")\n",
        "print(\"  JSON:    ~16.4 MB  (ethno_dataset_v2.json)\")\n",
        "print(\"  Parquet: ~761 KB   (ethno_dataset_v2.parquet, Snappy-compressed)\")\n",
        "print(\"  Ratio:   ~22× smaller\")"
      ]
    },
    {
      "cell_type": "markdown",
      "id": "c7870648",
      "metadata": {},
      "source": [
        "---\n",
        "\n",
        "## Get the Full Dataset\n",
        "\n",
        "| | Free Sample | Full Dataset |\n",
        "|---|---|---|\n",
        "| **Records** | 400 | 104,388 |\n",
        "| **Schema** | 8 columns | 8 columns |\n",
        "| **Enrichment fields** | Placeholder values | Real values (CT, ChEMBL, Patents) |\n",
        "| **Price** | Free | €699 one-time |\n",
        "\n",
        "**Purchase:** [ethno-api.com](https://ethno-api.com)  \n",
        "**Questions:** founder@ethno-api.com"
      ]
    }
  ]
}