A log of progress for the week of September 22, 2025.
This week marked a significant leap in the project’s capabilities, moving beyond core features and into the realm of Artificial Intelligence. After shipping a few final enhancements to the properties feature, the main challenge was to solve a critical business problem: how to automatically extract data from official documents. The solution involved building a custom AI-powered pipeline from the ground up.
Finalizing Property Features
Before diving into the AI work, I shipped several enhancements to the property management features. This included a new API endpoint for fetching property counts, the ability to filter properties by their ‘kind’, and updates to the UI to display these new statistics to the landlord. These updates rounded out the core feature set for properties, making the dashboard more informative.
Building an AI-Powered OCR Pipeline
The biggest challenge of the week—and perhaps the project so far—was automating the processing of uploaded Ejari PDF contracts. The standard AWS service for this, Textract, is not available in the UAE (me-central-1) region. This required a creative solution.
- The Solution: I designed and built a custom OCR (Optical Character Recognition) pipeline. This serverless infrastructure takes an uploaded PDF, processes it, and uses a Large Language Model (LLM) via AWS Bedrock to intelligently identify and extract the key information—like names, dates, and rental amounts.
- Implementation: The final days of the month were dedicated to implementing this pipeline. By the end of the day on the 30th, the feature was live: users could upload an Ejari PDF, and the system would automatically parse and populate the data, saving immense manual effort.
This week was transformative. By leveraging the power of generative AI with AWS Bedrock, we overcame a regional service limitation and built one of the application’s most valuable features. The project now has a powerful AI-driven capability, opening up many new possibilities for the future.