Refocus ag-forecasting page on OpenLambda, not the app domain

Trim crop/weather/model detail; keep why it's a useful test workload
and what the port drove in OpenLambda.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: mariaob1201

applications/ag-forecasting-api.md

@@ -1,58 +1,37 @@
 ---
 myst:
   html_meta:
-    description: "FastAPI/ASGI crop disease forecasting service from UW–Madison DSI, ported to OpenLambda."
-    keywords: "AgForecast, FastAPI, ASGI, OpenLambda, crop disease, Wisconet, serverless"
+    description: "A FastAPI/ASGI application used as a case study for porting real workloads to OpenLambda."
+    keywords: "OpenLambda, serverless, FastAPI, ASGI, case study"
 ---
 
 # Agricultural Forecasting API
 
-A FastAPI-based ASGI service for crop disease risk forecasting using multi-source weather
-data, developed by the [University of Wisconsin–Madison Data Science Institute](https://dsi.wisc.edu/)
-and ported to OpenLambda as a case study in deploying real-world applications.
+A FastAPI-based ASGI service from the [UW–Madison Data Science Institute](https://dsi.wisc.edu/),
+used here as a **case study for OpenLambda** — an existing, real-world application that was
+never designed for serverless, ported to the platform to find out what OpenLambda needs.
 
----
-
-## Overview
-
-The Ag Forecasting API provides geospatial agricultural intelligence for Wisconsin,
-combining weather data with peer-reviewed crop disease forecasting models. It exposes two
-parallel data pipelines through a unified interface:
-
-- **Wisconet** — the public mesonet of weather stations across the state.
-- **IBM Environmental Intelligence** — point-location queries by latitude and longitude.
+```{note}
+This page focuses on what the port taught us about OpenLambda. For the application itself,
+see its [source repository](https://github.com/UW-Madison-DSI/ag_forecasting_api).
+```
 
-The core logic lives in the `ag_models_wrappers` module, which dynamically pulls the daily
-and hourly weather variables each model requires for a given forecasting date, runs the risk
-calculations, and returns localized predictions.
+## Why this application
 
-## Supported models
+It's a good stress test for OpenLambda because it is non-trivial and ordinary: a FastAPI
+(ASGI) backend with real package dependencies, weather-data caching, and parallel compute —
+the kind of app teams actually run, not a serverless demo.
 
-All models are based on peer-reviewed plant pathology research from UW–Madison:
+## What it drove in OpenLambda
 
-- **Sporecaster** — white mold in soybean (dry and irrigated row-spacing variants)
-- **Tarspotter** — tar spot of corn
-- **Gray leaf spot** (corn)
-- **Frogeye leaf spot** (soybean)
+Porting it surfaced five concrete challenges that drove four new OpenLambda features:
 
-## Architecture
+- **Per-function environment variables** — to redirect writes away from read-only directories.
+- **`pip-compile` as a lambda** — to pin dependencies inside the OpenLambda environment.
+- **Direct GitHub deployment** — `ol admin install` straight from a repo URL.
+- **Built-in ASGI support** — OpenLambda acts as the ASGI server for FastAPI/Starlette/etc.
 
-The service is built on FastAPI (ASGI), with a Starlette `WSGIMiddleware` wrapper so it can
-also be served behind WSGI servers for legacy or mixed environments. A companion sub-package,
-`pywisconet`, provides a thin REST wrapper over the Wisconet v1 API for active-station
-discovery, station field metadata, and bulk measurement retrieval. IBM credentials are
-supplied via environment variables (`IBM_API_KEY`, `TENANT_ID`, `ORG_ID`).
+It also pointed to a future feature: `/dev/shm` support so process pools work under OpenLambda.
 
-## Running on OpenLambda
-
-Porting AgForecast surfaced five concrete challenges — writable directories, package version
-pinning, GitHub deployment, asynchronous execution, and parallel pools — that drove four new
-OpenLambda features: per-function environment variables, `pip-compile` as a lambda, direct
-GitHub deployment, and built-in ASGI support. The full write-up is in the
+The full porting write-up is in the
 [blog case study](../blog/post/2026-05-18-ag-forecasting-case-study.md).
-
-## References
-
-- Source: [UW-Madison-DSI/ag_forecasting_api](https://github.com/UW-Madison-DSI/ag_forecasting_api)
-- Live deployment: [connect.doit.wisc.edu/ag_forecasting_api](https://connect.doit.wisc.edu/ag_forecasting_api)
-- License: MIT