Skip to content

Getting Started

Three paths in. Each takes about five minutes. Pick the one that matches what you want; none of them requires the others.

The public dashboard is rebuilt every day and hosted on this site:

https://senthilsweb.github.io/ai-agents/trends/

Open it, filter, click a table row for posting details. The ? icon at the top right explains every chart in plain words. To track your own role keywords, add them to the URL — …/trends/?roles=ai engineer,engineering manager — and the page shows a live tracker for exactly those roles (the public build embeds no keywords of its own).

To build your own copy from today's live postings, with full job descriptions embedded (Docker only):

docker run --rm -v "$PWD/out:/app/exports" ghcr.io/senthilsweb/job-scout trends
open out/hiring-trends-*.html

The trends job fetches every configured job board (a few minutes), exports a parquet snapshot, and renders the dashboard into the mounted out/ folder.

Path 2 — Query the public dataset (needs only DuckDB)

At the end you will have live hiring data in a SQL prompt. No download, no account, no checkout.

-- in any DuckDB shell (duckdb from Homebrew/pip works)
SELECT category, count(*) AS jobs
FROM 'https://raw.githubusercontent.com/senthilsweb/ai-agents/main/agents/job-scout/data/ats_raw_trends.parquet'
GROUP BY 1 ORDER BY 2 DESC;

The file is refreshed every day at 11:00 UTC. Column meanings and more example queries: data/README.md. Prefer a browser? The hosted SQL console at https://senthilsweb.github.io/ai-agents/console/ runs DuckDB in your browser — its default query already reads this dataset.

Path 3 — Run the notebook locally (full pipeline)

At the end you will have a ranked, scored shortlist of postings in an interactive notebook, backed by a local DuckDB file.

cd agents/job-scout
pip install marimo duckdb pyyaml pandas python-dotenv anthropic certifi
marimo edit notebook.py

Then, inside the notebook:

  1. The database schema is created automatically. Every table starts empty.
  2. Click the "Fetch postings via ATS APIs (Tier 1)" button. It seeds companies from config.yaml and pulls live postings that match your title keywords.
  3. Re-run the "Ranked board" cell (or "Run all cells"). It does not refresh by itself after a fetch — this is a marimo behavior, see the Runbook.
  4. Move the scoring sliders (domain fit, compensation, and so on). This part is reactive and recomputes instantly.

Your shortlist is the ranked board. CSV and parquet copies land in exports/ every time the exports cell runs.

What next