SENTIMENT ANALYSIS

Paste the reviews. See what they're really saying.

Customer sentiment scored, themed, and summarized. From noise to signal.

REVIEW INPUT

Try a pre-loaded sample, or paste your own reviews below.

SENTIMENT DASHBOARD

Paste reviews or pick a sample set to analyze the sentiment.

Behind the Build
Stack
Cloudflare WorkersWorkers AI (NLP)Review source connectors (Google / Yelp / Trustpilot / internal DB)Workers Queues (batch)D1 (history)Vectorize (theme clustering)Slack + email alertsCRM webhooks (HubSpot / Salesforce)
Architecture
Review sources
Ingestion + dedupe
Per-review scoring
Theme clustering
Threshold alerts + CRM sync
Dashboard
How it works

A scheduled ingestion worker pulls reviews from connected sources (Google, Yelp, Trustpilot, internal feedback DB) and dedupes into a queue. Each review is scored for sentiment and topic-tagged via an LLM pass. Themes are clustered across the batch using embeddings so recurring issues surface even with different wording. An aggregate rollup hits a dashboard; spikes in negative sentiment or new critical themes trigger Slack and email alerts, and at-risk customer reviews can fire webhooks to CRM for an owner to follow up.

What you are seeing

Three pre-loaded review sets for Trailhead Outdoor Co., a fictional outdoor gear retailer, plus a free-text paste box wired to Workers AI (Gemma 4 26B with Llama 3.1 8B fallback). Results render as a donut chart, theme cards, highlight reel, and recommended actions. The demo stops short of live review-source connectors and alert fan-out — those are the production-only pieces above.

Try Meeting Intelligence to see sentiment extracted from live conversations.

More where that came from.

Back to all demos →