Hacker News

Subscribe to Hacker News feed
Hacker News RSS
Updated: 1 hour 31 min ago

Golden Age of Web Design

Wed, 02/18/2026 - 8:47pm
Categories: Hacker News

PicoLM: Run a 1B parameter LLM on a $10 board

Wed, 02/18/2026 - 8:38pm
Categories: Hacker News

Show HN: Analytics that tells AI product teams where their AI fails user

Wed, 02/18/2026 - 8:02pm

Traditional analytics tracks clicks. For AI products, you need to know: what was the user trying to do, did the AI help, and did they succeed?

I built a demo of this. It ingests AI conversations and runs 3 workers (GPT-4o-mini): intent classifier, quality scorer (LLM-as-judge), and task completion detector. Results show up in a dashboard designed for PMs, not engineers.

Stack: Python SDK (zero deps, async) → FastAPI → Supabase → GPT-4o-mini workers → Next.js dashboard.

Demo with sample data (not live product, validating the concept): https://dashboard-xi-taupe-75.vercel.app

The sample data models an AI app builder. Interesting patterns: scaffolding works great (78% success), but API integrations fail 75% of the time, and users who enter bug-fix loops almost always churn.

Key design question: is the "insights layer" (auto-generated recommendations, revenue-at-risk estimates, root cause identification) valuable enough to differentiate from Langfuse/Helicone adding product analytics to their existing tracing tools?

Looking for honest feedback, especially from AI product builders.

Comments URL: https://news.ycombinator.com/item?id=47068605

Points: 1

# Comments: 0

Categories: Hacker News

Will I Be Paid in Tokens?

Wed, 02/18/2026 - 8:01pm
Categories: Hacker News

Pages