🌱 Verdant

AI-powered daily garden insights from live sensor data — compared across Claude and GPT models, published as a static blog.

A .NET 8 pipeline reads soil moisture, temperature, and humidity from an Ecowitt sensor network, queries multiple AI models in parallel, and commits a Markdown comparison post to the repo. GitHub Actions then builds the Astro site and deploys it to GitHub Pages — so every morning a fresh post appears with no manual work.

How It Works

Ecowitt sensors
      │
      ▼
.NET 8 pipeline ──► Anthropic API  (claude-opus-4-6, claude-sonnet-4-6)
      │         └──► OpenAI API    (gpt-4o, gpt-4o-mini)
      │                    │ (parallel)
      ▼                    ▼
 Markdown comparison post  ←────────────────┘
      │
      ▼
 git commit/push ──► GitHub Actions build ──► GitHub Pages

Sensors (Ecowitt)

• GW1200 gateway — local Wi-Fi hub that polls all sensors every 5 minutes
• WH51 soil moisture sensors — one per bed
• WN32 outdoor temperature/humidity

Pipeline (pipeline/)

File	Purpose
Models.cs	All shared records + IInsightGenerator interface
GardenInsightParser.cs	Shared prompt builder and response parser
AnthropicInsightGenerator.cs	Calls the Anthropic API
OpenAIInsightGenerator.cs	Calls the OpenAI API
MockSensorProvider.cs	Simulates Ecowitt data; swap for real client when ready
BlogPostPublisher.cs	Renders a per-model comparison post as Markdown
Program.cs	Orchestrates the full pipeline

Site (site/)

• Astro v4 static site with a content collection for posts
• Minimal, readable layout with sensor data tables and per-model sections
• Deployed to GitHub Pages via the Actions workflow

Repo Structure

verdant/
├── pipeline/
│   ├── GardenAI.csproj
│   ├── Program.cs
│   ├── Models.cs
│   ├── GardenInsightParser.cs
│   ├── AnthropicInsightGenerator.cs
│   ├── OpenAIInsightGenerator.cs
│   ├── MockSensorProvider.cs          ← swap for real EcowittClient here
│   └── BlogPostPublisher.cs
├── site/
│   ├── package.json
│   ├── astro.config.mjs               ← set your GitHub Pages URL here
│   ├── tsconfig.json
│   └── src/
│       ├── content/
│       │   ├── config.ts              ← Astro content collection schema
│       │   └── posts/                 ← generated .md files land here
│       ├── layouts/
│       │   ├── Base.astro
│       │   └── PostLayout.astro
│       ├── pages/
│       │   ├── index.astro
│       │   └── posts/[...slug].astro
│       └── styles/
│           └── global.css
├── .github/
│   └── workflows/
│       └── daily-pipeline.yml
├── .gitignore
└── README.md

Setup

1. GitHub Repository Secrets

Go to Settings → Secrets and variables → Actions and add:

Secret	Value
ANTHROPIC_API_KEY	Your key from console.anthropic.com (sk-ant-...)
OPENAI_API_KEY	Your key from platform.openai.com (sk-...)

GITHUB_TOKEN (for the bot commit and Pages deploy) is provided automatically.

2. Enable GitHub Pages

1. Go to Settings → Pages
2. Set Source to GitHub Actions (not "Deploy from a branch")
3. That's it — the workflow handles the rest

3. Update site/astro.config.mjs

Edit the two lines at the top of the config to match your repository:

// For a project repo at https://username.github.io/verdant/
site: 'https://username.github.io',
base: '/verdant',

// For a user/org site at https://username.github.io/
site: 'https://username.github.io',
// (remove the base line)

// For a custom domain at https://garden.example.com/
site: 'https://garden.example.com',
// (remove the base line, and add a CNAME file to site/public/)

Commit this change before the first pipeline run.

4. Local Development

Run the pipeline (writes a post into the site's content directory):

export ANTHROPIC_API_KEY=sk-ant-...
export OPENAI_API_KEY=sk-...
export POSTS_OUTPUT_DIR=./site/src/content/posts
dotnet run --project pipeline/

Preview the Astro site:

cd site
npm install
npm run dev        # http://localhost:4321

Build the static site:

cd site && npm run build   # output in site/dist/

5. Adding or Removing Models

Edit the generators array in pipeline/Program.cs:

IInsightGenerator[] generators =
[
    new AnthropicInsightGenerator(anthropicApiKey, "claude-opus-4-6"),
    new AnthropicInsightGenerator(anthropicApiKey, "claude-sonnet-4-6"),
    new OpenAIInsightGenerator(openAiApiKey, "gpt-4o"),
    new OpenAIInsightGenerator(openAiApiKey, "gpt-4o-mini"),
];

All models are queried in parallel; the post includes one labelled section per model. If a model fails its API call, it is noted in the post but doesn't block publication.

Switching to Real Ecowitt Data

pipeline/MockSensorProvider.cs simulates Ecowitt readings. When you're ready for live data, replace its two methods with calls to your GW1200's local API:

1. Find your gateway's local IP (Ecowitt app → Settings → Network)
2. Call GET http://<gateway-ip>/api/v1/device/real_time?application_key=...&api_key=...
3. Map the response JSON onto SensorReading and DailyHistory in Models.cs

Everything downstream (AI generation, post rendering, git commit) is unchanged.

Schedule

The pipeline runs at 8:00 am NZST every day (20:00 UTC).

Manual run: Actions → Daily Garden AI Pipeline → Run workflow.

Tech Stack

Layer	Technology
Sensors	Ecowitt GW1200, WH51, WN32
Pipeline	.NET 8, C#
AI — Anthropic	claude-opus-4-6, claude-sonnet-4-6
AI — OpenAI	gpt-4o, gpt-4o-mini
Blog	Astro v4
Hosting	GitHub Pages
CI/CD	GitHub Actions