Sentiment Spark
An AI sentiment-analysis tool for pasted text or any web-page URL - scoring, confidence, sentence-level breakdown, keywords, and a summary, in one pass.
The project, quickly
Solo developer - frontend and serverless backend in a single repository.
Vite, React, TypeScript, Tailwind CSS, and shadcn/ui.
Two Supabase Edge Functions - one for scraping, one for analysis.
Built and functional, not yet deployed to a public URL.
The problem
Reading sentiment out of a long review, article, or thread manually is slow, and most sentiment tools only take pasted text - if the thing you care about lives at a URL, you have to copy it out yourself first.
Sentiment Spark takes either: paste text directly, or hand it a URL and let it scrape the content itself. Either way, it comes back with more than just "positive" or "negative" - a confidence score, a sentence-level breakdown, extracted keywords, and a short summary.
How it works
URL scraping
A scrape-url Supabase Edge Function calls Firecrawl to pull the main
content from a page - YouTube, Reddit, blogs, and news are all supported.
AI sentiment analysis
The extracted or pasted text is sent to an analyze-sentiment Edge
Function, which calls an AI gateway and returns a structured JSON result.
Sentence-level detail
Beyond a single score, the result breaks sentiment down sentence by sentence, alongside extracted keywords and insights.
Sentiment gauge & summary
The UI renders a sentiment gauge plus a concise written summary, so the result is legible at a glance and in detail.
Tech stack
Where it stands
The core loop works end to end - scrape or paste, analyze, display - but it's sitting behind
local environment variables rather than a public deployment right now. Getting it live just
means wiring up the Supabase secrets and API keys in a hosted environment rather than a
local .env file.