Deploying the LLM Grader on Render

Overview

LLM grader is generally hosted on a server. We suggest Render that can directly load from a GitHub repository and a persistent disk. Render handles the build, environment, and hosting automatically, so deployment is simple and repeatable.

A Render deployment consists of:

  • a Web Service running the Flask application
  • a Persistent Disk mounted at /var/data
  • environment variables for Google sign-in and admin authorization
  • automatic redeploys on Git pushes

The grader stores uploaded solution packages on the persistent disk, so course content survives restarts and redeploys. Before starting you will need to create a Render account by linking your GitHub account to Render.

🧱 1. Create a Persistent Disk on Render

In the Render dashboard:

  1. Go to Disks
  2. Click New Disk
  3. Name it something like grader-data
  4. Choose a size (1–2 GB is plenty)
  5. Set the mount path to:
/var/data

This directory will store:

  • extracted solution packages
  • logs (if you choose to write any)
  • future assets such as images

Render guarantees that /var/data persists across deploys.

🌐 2. Create the Web Service

  1. Click New → Web Service
  2. When prompted for a repository, select your fork of llmgrader, not the original upstream repository
  3. Choose the branch in your fork that you want Render to deploy, typically main
  4. Use these settings:

Environment:

Python 3.x

Build Command:

pip install -r requirements.txt

Start Command:

gunicorn run:app

Grading now runs as a background job with polling, so long grade operations no longer depend on a single long-lived /grade request. The current implementation keeps grading simple and safe by allowing only one active grading job per app instance at a time.

Instance Type:

  • Start with Starter or Basic
  • Upgrade later if needed

Persistent Disk:

  • Attach the disk you created
  • Mount it at /var/data

Render will clone and deploy from the fork and branch you selected above. That lets you maintain an independent version of the repo for your own portal.

🔐 3. Set Environment Variables

In the Render dashboard:

  1. Open your web service.
  2. Go to Environment.
  3. Find the Environment Variables section.
  4. Click Add Environment Variable.
  5. Enter the variable name in the Key field and its value in the Value field.
  6. Repeat for each variable below.
  7. Click Save Changes when you are done.

Add these variables:

Variable Suggested value Remarks
LLMGRADER_SECRET_KEY Long random string Stable Flask session secret
LLMGRADER_GOOGLE_CLIENT_ID OAuth client ID Google OAuth configuration
LLMGRADER_GOOGLE_CLIENT_SECRET OAuth client secret Google OAuth configuration
LLMGRADER_GOOGLE_REDIRECT_URI https://<host>/auth/callback OAuth callback URL
LLMGRADER_INITIAL_ADMIN_EMAIL you@example.com Bootstraps first admin
PYTHON_VERSION 3.12.3 Update with python version
LLMGRADER_STORAGE_PATH /var/data/ Root for persistent storage
LLMGRADER_PRIVATE_KEY (generated value) Optional — required only if using submission signing

Example values for a Render deployment might look like this:

LLMGRADER_SECRET_KEY = <a long random string that you generated>
LLMGRADER_GOOGLE_CLIENT_ID = 123456789012-abcdefg.apps.googleusercontent.com
LLMGRADER_GOOGLE_CLIENT_SECRET = GOCSPX-xxxxxxxxxxxxxxxx
LLMGRADER_GOOGLE_REDIRECT_URI = https://your-service-name.onrender.com/auth/callback
LLMGRADER_INITIAL_ADMIN_EMAIL = you@example.com
PYTHON_VERSION = 3.12.3
LLMGRADER_STORAGE_PATH = /var/data
LLMGRADER_PRIVATE_KEY = <base64-encoded private key from generate_signing_keys>

For more detail on where the Google OAuth values come from, how to choose LLMGRADER_SECRET_KEY, and how to set the redirect URI, see Setting up Google OAuth.

Notes:

  • LLMGRADER_GOOGLE_REDIRECT_URI must exactly match one of the authorized redirect URIs in Google Cloud.
  • LLMGRADER_STORAGE_PATH should match the persistent disk mount path.
  • LLMGRADER_SECRET_KEY is your app’s Flask session secret, not a Google value.
  • After saving environment-variable changes, Render will usually trigger a redeploy. If it does not, manually redeploy the service.

🔄 4. Deploy and Verify

Render will:

  • clone the selected branch from your fork
  • install dependencies
  • start the app with Gunicorn
  • mount the persistent disk

Once deployed, render assign your service a unique URL such as:

https://llmgrader-xxxx.onrender.com

Use the URL shown in your Render dashboard.

🛠️ 5. Redeploying

Render redeploys automatically when you push to the branch that your service is tracking in your fork.
You can also trigger a manual deploy from the dashboard.

Uploads and course content remain intact because they live on the persistent disk.

🧹 6. Cleaning or Resetting the Disk (Optional)

If you ever need to reset the grader:

  • SSH into the instance (Render Shell)
  • Remove the contents of /var/data/soln_repo
  • Or delete and recreate the disk from the dashboard

This does not affect your code deployment.

Once you are deployed, you can upload the course package

Go to Google sign-in and admin access

This site uses Just the Docs, a documentation theme for Jekyll.