🍽️ ReviewPulse – Automated Restaurant Ratings & Reviews Monitor

End-to-end automation built with n8n, Google Sheets, and OpenAI, designed to track restaurant reputation changes at scale.

✨ What it does

• 🔄 Daily updates (~1,300 restaurants, split into 5 groups to save API calls)
  
• 📝 Change logging → Only rating/review changes are appended to a “Deltas” sheet
  
• 📩 Digest reports → Every 5 days, changes are summarized into a report and emailed automatically
  
• 🤖 AI summary (optional) → Generate highlights like “Top movers” or “Biggest shifts”
  

🚀 Workflow Architecture

1. Daily Workflow

• ⏰ Triggered by schedule (runs every day)
  
• 🗂️ Reads full restaurant list and splits into groups (~260/day)
  
• 🔍 Pulls fresh ratings & reviews (Google Places API)
  
• 📊 Merges with existing data → calculates deltas
  
• ✅ If something changed → appends to Deltas sheet
  

2. Digest Workflow (every 5 days)

• ⏰ Triggered every 5th day
  
• 📥 Reads all entries from Deltas sheet for the last 5 days
  
• 📊 Summarizes key changes (biggest rating shifts, most new reviews)
  
• 🤖 (Optional) Sends the summary through OpenAI for a polished report
  
• 📧 Delivers via Gmail
  

🛠️ Tech Stack

• n8n (self-hosted) → Orchestration, scheduling, error handling
  
• Google Sheets → Source of truth & append-only change log
  
• Google Places API → Fetches ratings and reviews
  
• OpenAI (optional) → Creates natural-language digests
  
• Gmail/SMTP → Sends automated reports
  

📊 Data Model

Restaurants (master list)
rownumber | name | address | placeid | rating | userRatingCount | lastchecked | runtoday | group

Deltas (append-only log)
date | placeid | rownumber | name | rating | userRatingCount | deltarating | deltareviews | changed

⚡ Challenges & Learnings

🔒 API Quotas & Limits

• Problem: Google Places API has strict daily request limits. With ~1,300 restaurants, calling all at once would burn through quota.
  
• Solution: Split the dataset into 5 rotating groups (~260/day) → only one group updates per run, keeping calls within limits.
  

🔀 Change Detection

• Problem: Pulling ratings daily would quickly fill sheets with duplicate data.
  
• Solution: Implemented a delta calculation step → only new or changed values (ratings/reviews) are logged.
  

📊 Data Merge Logic

• Problem: Combining new API results with the existing dataset while keeping row order and metadata consistent.
  
• Solution: Built a merge-and-calculate step in n8n → ensures updated ratings/reviews are written in place, while old data persists.
  

⏱️ Digest Scheduling

• Problem: If emails were sent daily, they would become noise and lose value.
  
• Solution: Introduced a 5-day batching workflow → aggregates all deltas before sending a digest, creating a useful “newsletter-style” report.
  

🧩 Workflow Structure

• Problem: A single massive workflow was harder to debug and maintain.
  
• Solution: Split into two modular workflows:
  
  • Daily updater (lightweight, API-focused)
    
  • 5-day reporter (summarization + email)
    

🛠️ Key Takeaways

• Respect API limits from day one
  
• Always log only the deltas
  
• Modular automation = easier debugging + better scalability
  
• AI summaries add polish, but the core automation works even without them
  

🚧 Next Steps / Improvements

• Add visual dashboards (Tableau/Looker Studio) to track long-term trends
  
• Integrate alerts (e.g., Slack/Telegram) for major rating drops
  
• Expand beyond restaurants → cafes, bars, or other local businesses
  
• Add multi-city support with a single consolidated pipeline
  

💡 Why this project?

I’m passionate about food 🍜, but I also wanted to build something that shows automation + data engineering skills:
- ✅ Automates real-world data collection
- 📈 Tracks changes without manual effort
- 🗞️ Produces clean, shareable digests
- 🔧 Shows modular workflow design and problem-solving