Applications

A bring-your-own-data, bring-your-own-LLM job application tracker backed by your own Google Sheet.

LM Studio Configuration

Self-hosted models are called straight from the browser. Nothing LLM-related is deployed with the app.

1. Load the model: in the Developer tab, load a chat model (7B+; smaller models often can't do structured output). In its load settings, set Context Length to 8192 or more, because the 2k/4k defaults cut off long postings. From the CLI: lms load <model> --context-length 8192.
2. Start the server: in the server settings, turn on Enable CORS and start the server.
3. Connect the app: in Settings → AI Provider, pick Self-hosted (OpenAI-compatible), set the URL to http://localhost:1234/v1/chat/completions, then click Discover and pick the model.
4. Leave the app's Structured Output toggle off. The JSON schema is sent with every request as response_format.
5. Check that the schema is enforced. Run the command below. It should return {"title": ...}, not a poem:

   curl http://localhost:1234/v1/chat/completions -H "Content-Type: application/json" -d '{
     "model": "<model id>",
     "messages": [{"role": "user", "content": "Write a poem about the sea."}],
     "response_format": {"type": "json_schema", "json_schema": {"name": "test", "strict": true,
       "schema": {"type": "object", "properties": {"title": {"type": "string"}},
                  "required": ["title"], "additionalProperties": false}}}}'

Key Features

• Applications in your Google Sheet: track applications, with status history stored in your own spreadsheet.
• AI auto-fill: fill the form from a pasted job posting with Gemini, OpenAI, Anthropic (via a CORS proxy) or any OpenAI-compatible server, with model discovery from the provider.
• Analytics: status and company breakdowns, a draggable pipeline Sankey, and a status timeline grouped by day, week or month.
• Printable report for any date range.
• Server-side logging: client logs are written to daily-rotated files by a log-sink container.

System Design

flowchart LR
  user["Browser (React SPA)"]
  subgraph host["Docker host"]
    traefik["Traefik"]
    app["applications<br/>nginx: SPA, /config.js, /api/logs proxy"]
    logger["applications-logger<br/>Node + winston"]
    logs[("LOG_DIR<br/>daily JSON logs")]
  end
  google["Google OAuth, Sheets,<br/>Drive, Picker APIs"]
  llm["LLM provider<br/>Gemini / OpenAI / Anthropic / LM Studio"]

  user -->|HTTPS| traefik --> app
  app -->|/api/logs| logger --> logs
  user -->|"read/write sheet"| google
  user -->|"extract posting"| llm

Loading

Details on logging, LLM requests and the time series are in .claude/DESIGN.md.

Local Dev Prerequisites

• Node.js >= 22 and npm
• Docker Compose >= 2.0 (production only)
• A Google Cloud project with the Sheets, Drive and Picker APIs enabled, an OAuth client ID and an API key. If the key has referrer restrictions, the app's origin must be allowed, and it must also be listed as an authorized JavaScript origin on the OAuth client.

Install dependencies with npm install, plus npm --prefix server install for the log sink.

Configuration & Environment Variables

Set these in .env (copy it from .env.example).

Variable	Default	Notes
VITE_GOOGLE_CLIENT_ID	—	Required. Baked into the bundle at build time
VITE_GOOGLE_API_KEY	—	Required. Baked into the bundle at build time
DOMAIN	apply.whitney.rip	Traefik host rule
LOG_LEVEL	info	debug/info/warn/error. Read at container start by both the client (via /config.js) and the sink
LOG_DIR	/pwspool/software/applications/logs	Host directory for log files
LOG_RETENTION	30d	Log files older than this are deleted
LOG_ROTATE_FREQUENCY	1d	How often a new log file is started
LOG_CONSOLE_FORMAT	pretty	json when a collector reads docker logs

Operational Runbook

# Local setup & development (http://127.0.0.1:5173)
git clone git@github.com:runyanjake/applications.git && cd applications
cp .env.example .env            # fill in the Google credentials
npm install
npm run dev
npm --prefix server install && npm run logs   # optional, second shell: log sink writing ./logs

# Linting & type-checking (no test suite yet)
npm run lint
npx tsc -b

# Production build & run (Jenkins runs the same steps, then health and smoke checks)
sudo mkdir -p /pwspool/software/applications/logs
docker build --target ci -t applications-ci .
docker compose up -d --build

# Common operations
docker logs -f applications-logger                                         # live log stream
tail -f /pwspool/software/applications/logs/applications-$(date -u +%F).log
jq -c 'select(.level == "error")' /pwspool/software/applications/logs/*.log
LOG_LEVEL=debug docker compose up -d                                        # change level, no rebuild
docker compose down