Collection from multiple sources
Separate adapters for Facebook and Khmer24, with error handling and controls for unnecessary traffic.
← All projects
APP UI RECONSTRUCTION
PROJECT / 02 / PERSONAL PRODUCT
Homeasy.
Turning scattered rental listings into one useful search
Data collectionTelegram mini appWorking prototype
APP UI RECONSTRUCTIONContext
The problem
Finding a rental in Cambodia can mean checking many Facebook posts and local sites. Listings use different formats and languages, and the same property may appear more than once.
Homeasy aims to bring those listings into one search and reduce the work of comparing details and tracking new posts.
Approach
A pipeline for each listing
I split the process into clear steps: collect a post, extract its details, normalize the fields, compare possible duplicates and send the result to the user interface.
01SourcesFacebook / Khmer24
→02ExtractionLanguage model + rules
→03MatchingDuplicates / quality
→04DeliveryAPI / bot / app
Simplified view of the listing pipeline.
What I built
Main product features
Structured data from text
LLM extraction and deterministic rules create a shared data model for filtering.
Duplicate detection
Compares listing details and image fingerprints (pHash) to reduce repeated results.
Search and alerts
Fastify API, React mini app, saved listings, filters and a Telegram bot with search alerts.
Result
A working end-to-end MVP
The product connects collection, processing, storage and search in a Telegram interface. I continue to improve data quality, speed and integration reliability.
The interactive example uses three fictional posts and combines one repeated listing. It demonstrates the rules in the demo; it is not a benchmark of the live parser or language model.
View selected code samples on GitHub ↗ — this is a public showcase, not the full production repository.
App screens
Search, filters and listing details.
Visual recreations based on the working app. All listings and contact details are synthetic.
Visual recreations of the live app interface; all listings and contact details are synthetic.