Jobwatch / openings at companies worth watching
Watch the careers pages without opening them
Jobwatch reads the public job-board endpoints of a hundred companies every morning and keeps every posting it has seen. A jury of three language models labels each one with the role family, the seniority, the workplace, the tools it names and the pay when the text states it. The companies are places I would work. They are health tech, sports, public-interest groups, product companies in the DC area and remote, and a few consumer brands. SQL runs in your browser. A question in words goes to a hosted model that writes the query.
The project page says how the warehouse is built and how the labels are checked. The Gamebot page (opens in a new tab) is the same machinery over a TV show.
Open data roles now
Every open posting the jury put in a data family, newest first. Each row links to the posting on the company's own board. The stack chips are the tools at least two jurors found in the text. Pay appears only when the posting states it.
Ask in words
A hosted model reads the tables, the label vocabularies and the column notes, and writes one DuckDB query for your question. The query lands in the editor below, where you can read it, change it and run it. The model never runs anything. It says what it assumed when the question left a choice open.
SQL
DuckDB runs in the page and reads each gold table's Parquet file on demand. The tables are gold.openings, gold.company_summary, gold.stack_month and the rest of the lineage map below. Clicking a table name in the sidebar writes a starter query. stack is a list, so list_contains(stack, 'databricks') or unnest(stack).
Did this answer the question? ·
Lineage
Bronze is what the boards said, one payload per posting per change. Silver is one row per posting with its open and closed dates, and one row per posting per labeler. Gold is what this page serves. Click a gold table for its columns and the SQL that builds it. The dashed tables stay in the warehouse and are not served here.
Companies
One card per company on the list, with what it has open now. A company without a public board is watched by hand and shows only its careers link. The alumni link opens the school's LinkedIn people page with the company as the keyword, which is where a warm introduction starts.
How the labels are made
Three models read each posting with the same prompt and answer from closed vocabularies. DeepSeek-Flash, GLM-5.3-Flash and GPT-5.6 Luna. The verdict is the majority, and GPT-5.6 decides a three-way split. A regex labeler reads the same text and is shown beside the vote without voting. The numbers below are agreement between labelers over every posting in the warehouse, and agreement is not accuracy. A label all three models get wrong the same way counts as unanimous.
Data: the public job-board endpoints of Greenhouse, Lever and Ashby, read once a day. Each posting is quoted here only as a short excerpt and linked to its source. No people data is collected. Built by Max Grody.