Sift

Every review, sorted by what to fix.

Type: AI web app
Year: 2026
Role: Design, development, launch video
Stack: Next.js 16, React 19, TypeScript, Tailwind CSS v4, OpenAI-compatible API

Paste customer feedback or upload a CSV. Sift groups it into themes, scores each one and puts the problem that hurts most at the top, with a next step.

Teams collect reviews, support tickets and survey answers, then never get through them. Sift turns that pile into a short, ranked list.

An AI model groups the responses into themes and scores sentiment. Each theme is ranked by how many people mention it, weighted by how unhappy they are, so the costliest one is pinned to the top with the worst real quote and one concrete action.

The demo runs on Bloomcart, a fictional plant shop. Names, reviews and numbers in the sample are made up, and the sample is pre-analyzed so the whole app can be explored without an API key.

Details worth a look

Fix-first card: the costliest theme, how often it comes up, and a recommended action
Weekly health score with the change since last week
Every response searchable and filterable, so you can check the grouping yourself
Paste text or import a CSV, up to 200 responses per analysis
Ctrl+K command palette to jump to any page or theme
Per-visitor and daily rate limits to keep a public demo's API bill small

Overview: the fix-first card, health score and sentiment split.

Theme detail: summary, recommended fix, weekly trend and every response in the theme.