Empoweromics Data Scientist 2022–2023

Lead scoring, message routing and smart search for a real-estate platform

Unified data from 3,008+ real-estate developers, built lead-prioritisation and WhatsApp-routing models, and shipped typo-tolerant client search.

See the architecture

Role: Data integration and ML models (Data Scientist) Full-time
Company: Empoweromics real-estate technology, Cairo
When: Jan 2022 – Mar 2023 1 yr 3 mos
Where: Cairo, Egypt
Stack: Python
PyTorch
NLP
Computer vision
Azure Functions
Cosmos DB
SQL Server

real-estate developers' data unified: 3,008+
client lifecycle stages the lead models predict: 7
departments that WhatsApp messages are routed to: 5

Fig. 1 empoweromics-message-routing · architecture

Figure 1 Automatic classification of WhatsApp messages (text and images) routes each message to the right department, alongside the ETL and data-integration work behind the company's real-time e-map. Diagrams show the components named in public descriptions of the work. They are simplified, not complete system maps.

Classification. Messages are classified automatically: text with NLP, images with computer vision.
Routing. Each message is routed to Tech, HR, Accounting, Operations or Marketing.
ETL. ETL from Cosmos DB (NoSQL) to Microsoft SQL Server, scheduled with Azure Functions triggers.
Data integration. Data from 3,008+ real-estate developers, each with its own schema, merged to scale the real-time e-map.

Text description of the diagram

Architecture diagram in two parts. Message routing: incoming WhatsApp messages are classified, text with an NLP model and images with a computer-vision model, and routed by department to Tech, HR, Accounting, Operations or Marketing. Data engineering: an ETL pipeline moves data from Cosmos DB (NoSQL) to Microsoft SQL Server on scheduled Azure triggers; a data-integration system merges data from 3,008+ real-estate developers, each with its own schema, to feed the real-time e-map.

On this page The problem

The problem

Empoweromics runs a real-time e-map for real estate. Its data came from 3,008+ developers, each with its own schema. Brokers needed to know which leads to chase, inbound WhatsApp messages (text and images) had to reach the right department, and staff had to find clients even when a name was misspelled.

Approach

A data integration system that merged 3,008+ developer schemas to scale the e-map.
Classification models for 7 client lifecycle stages (Inactive, Assigned, In Progress, Lost, Won, Void, Active) to rank leads for brokers.
WhatsApp classification with NLP for text and computer vision for images, routing to Tech, HR, Accounting, Operations or Marketing.
Smart Search: fuzzy matching over Arabic and English names, robust to misspelled or partial input.
A PyTorch model that infers gender from Arabic names, for targeting.
ETL from Cosmos DB to SQL Server, scheduled with Azure Functions.

Architecture

Component view: developer data is unified for the e-map; ETL moves data from Cosmos DB to SQL Server; separate models serve brokers, campaigns, client lookup and message routing. Figure 1

Users and apps: Real-time e-map
Brokers
Targeting and campaigns
WhatsApp inbox
Models: Lifecycle-stage models
Name-to-gender model (PyTorch)
Text + image classifier
Retrieval: Smart Search (fuzzy)
Data: 3,008+ developer data sources
Cosmos DB
SQL Server
Client records
Services: Department queues
Pipelines: Schema integration
ETL (Azure Functions)
Connections: 3,008+ developer data sources to Schema integration
Schema integration to Real-time e-map
Cosmos DB to ETL (Azure Functions)
ETL (Azure Functions) to SQL Server
Lifecycle-stage models to Brokers (lead priority)
Name-to-gender model (PyTorch) to Targeting and campaigns
Smart Search (fuzzy) to Client records
WhatsApp inbox to Text + image classifier
Text + image classifier to Department queues

Outcome

Data from 3,008+ developers unified so the real-time e-map could scale.
Brokers prioritise leads by predicted lifecycle stage; inbound WhatsApp messages reach the right department automatically.

Stack

Models: PyTorchNLPComputer vision
Data: Cosmos DBSQL Server
Services: Python
Pipelines: Azure Functions

Where these facts come from. Everything on this page comes from my CV, my LinkedIn profile and my GitHub profile and repositories. Nothing is estimated: where no figure is public, the page describes what the system does.

Next case study

Logic Controls2026

logic-controls.com: a 4-language site, accessible and tested end to end

Freelance build of logic-controls.com in English, Arabic (RTL), Chinese and Turkish: Astro on Cloudflare Workers, WCAG 2.2 AA, Lighthouse 98–100 mobile.

Astro
TypeScript
Tailwind CSS
Cloudflare Workers
Turnstile

Questions about this build?

Ask my AI about “Real-estate ML at Empoweromics”.

It answers from my CV and public profile, in English or Arabic.

Contact me or write to hatem@mohamedhatem.dev