Ho Jiak Bo: a food recommender app based on top food blogs

Build an AI food recommender by crawling Singapore blogs, extracting JSON via LLMs, adding map data, and blending full‑text with embedding search.

Overview

Ho Jiak Bo is an end-to-end, AI-powered recommender that ingests and unifies Singapore’s top food-blog reviews into a single, searchable interface. In my talk, I’ll walk through:

Data ingestion pipeline – using Python + asyncio and Playwright to crawl and render JavaScript-heavy blogs (Daniel Food Diary, EatBook.sg, Miss Tam Chiak).
Content structuring – how we leverage DeepSeek/OpenAI to extract clean JSON records, enrich them with Google Maps metadata, and generate embeddings.
Hybrid search engine – combining PostgreSQL full-text search and OpenAI embeddings (text-embedding-3-small) with distance-decay scoring (via PostGIS) to deliver top-5, location-aware hawker stall recommendations.

Links

Tech stack