Filter AI news with three tuned questions

01 · Role in the system

What Jev does here

No LLM agent and no code nodes: You.com searches each topic's news from the past day, seen articles are skipped, and one Jev call per article asks whether it reports one specific new development (yes/no, passes at 0.8), whether it is about the topic (yes/no, 0.8) and how much it matters to the team (0-2 score, passes at 1.3). Material articles group by topic, You.com Research writes one cited brief per topic, Slack gets the post; a morning costs $0.01-0.04 and topics with no material news post nothing. The tuning is the lesson: v1 asked only on-topic and importance and flagged an encyclopedia page, two funding-database pages and an explainer - adding the narrow 'single new development' question with counter-examples fixed it. Thresholds were set on four weeks of 674 articles (160 hand-labeled blind) and checked on the four weeks held out.

02 · Control boundary

Where Jev sits

Retrieval, typed decisions and routing as three pluggable halves: You.com fetches, the official TypeSafe AI n8n node (jev-latest) judges three questions per article in one call, and thresholds in code decide what survives to grouping and briefing - no agent loop anywhere.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• News briefs depend on You.com's Research API for wording and citations; Jev only selects and groups.
• Thresholds are team-specific (materiality especially) and were tuned for AI-industry topics.
• The workflow is vendor-published by You.com's OSS arm; both engines are commercial APIs.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

• youdotcom-oss ↗Community · github · public · checked 2026-10-17