This content originally appeared on DEV Community and was authored by Michał Śnieżyński
This is a submission for the AI Agents Challenge powered by n8n and Bright Data
What I Built
This AI agent provides observability for app store reviews and delivers valuable insights.
It analyzes the sentiment of each review and stores reviews in external data storage.
It allows you to send Slack notifications, so you can get notified about any new reviews.
This can be extremely useful when you want to react quickly to new negative or neutral reviews, or simply to get real-time ideas about what you should improve – for example, after a version release.
Demo
Example prompt should contain URL address to app in AppStore (any market, without a query param):
https://apps.apple.com/us/app/messenger/id454638411
send reviews insights to Slack
Example output:
7 new positive reviews
0 new neutral reviews
3 new negative reviews
Insights:
• Spam/scam exposure — Harms trust; tighten filters/reporting (20% cite spam/scams)
• Call audio issues — Hurts call quality; improve noise/volume (20% mention crackle/static)
• Privacy controls — Clarify recording opt-in/out (1 user fears "Big Brother")
• Volume too low — Accessibility risk; add boost/test (1 report: "volume not loud enough")
• Overall satisfaction — Maintain core UX (70% positive reviews)
n8n Workflow
Technical Implementation
I used XML for the Agent’s system instructions.
For sentiment analysis, I used the cheap GPT-5-nano, as it doesn’t require a better model.
For insights and the AI Agent, I decided to use GPT-5/GPT-5-mini – usually for those kinds of tasks, bigger models perform better.
For memory, I kept it simple with short chat history storage.
For data storage, I used Supabase PostgreSQL.
Bright Data Verified Node
I used the Bright Data node to scrape reviews from App Store. I also tested the node to get SERPs. Working with the node is smooth. The only thing I would improve is adding a limit option for scrapers.
Journey
At first, I wanted to build one AI Agent that did everything. But this was my first time using n8n, and I had trouble getting the agent to work with its tools.
Since I know some automation stuff, I decided the scraping part needed to be more reliable. So I made a separate workflow for scraping, sentiment analysis, and generating summaries.
I kept the AI Agent extendable, so users can add more sub-workflows for other marketplaces.
This content originally appeared on DEV Community and was authored by Michał Śnieżyński