I built a pipeline that tells me what to build next

Every morning I was making the same guess: what search trend is big enough right now that I could ship a tool around it in a weekend? Gut feel is a bad input. So I built something that answers the question with data instead.

This morning it found four clusters independently: spring break travel, community clubs, third places, making friends. Nine keywords hit breakout simultaneously across three separate sources. The pipeline named the through-line: social connection. Multiple search angles on the same underlying need. I wouldn't have connected those manually — I would have seen four separate trends and picked one at random.

What shipped

• Daily briefing, automated. Every morning at 9AM the pipeline runs without me touching it. I open briefings/briefing_YYYY-MM-DD.md and read a ranked list of trend clusters with scores, breakout signals, and a plain-English "my read" on what to build.
• Three data sources, cross-referenced. trendspy (Google's internal API — 400+ trending topics per call), Google Trends RSS, and Gmail newsletter ingest all run in parallel. A keyword appearing in 2+ sources gets a confidence boost. One source = a trend might be happening. Two sources = it's real.
• Scoring that actually means something. Each keyword gets a 0–100 score: 35% growth velocity, 25% buildability (does the search query imply a tool someone would pay for?), 20% volume, 20% freshness from 30-day time series data.
• Reddit pain signal validation. The top clusters get searched on Reddit automatically. The system flags posts with "wish there was", "looking for", "frustrated" — and marks results as unreliable if the pain posts are clearly off-topic. No false confidence.
• The whole thing runs ~13 API calls. Free tier Brave Search for competitor checking, OpenAI for cluster naming. Everything else is free.

Why

Before this, the decision was Twitter, Reddit, and whatever felt interesting on a given morning. An hour of reading and I'd still end up on gut feel. The data existed — Google publishes it — it just needed pulling together and scored against what's actually buildable in 48 hours.

I built this for me first. I'm the one who needs to know what to build next.

What's next

Run it privately for a few months to validate signal quality. If the output holds, it becomes a public tool: enter a niche, get back the five trends most worth building around this week.

The issues are parked until then.