Skip to content
Karim Alaywan

Work

Analysis built around a decision.

One system I built and run, and five projects across revenue management, customer analytics, healthcare, data engineering and applied AI.

Flagship project

Commercial Intelligence OS

A system for B2B lead discovery, prioritisation and sales decision support, built on PostgreSQL, Supabase, PostGIS and n8n.

View project

Machine Learning × Revenue Management

Predicting Hotel Cancellations

Hotels lose revenue to last-minute cancellations, but overbooking blind gets guests walked. Predict each reservation, then price the two errors against each other.

$10,620

Additional modelled profit from moving the decision threshold to 0.55. A simulated result on a public dataset, not realised revenue.

Python · XGBoost · SMOTE · Poisson simulation

Customer Analytics × Predictive Modeling

Predicting Digital Disengagement

In a multi-SIM market, a silent customer may be switching handsets or may already be gone. The operator needs to tell those apart before revenue drops.

Built as an early-warning problem rather than a scoring exercise: the model was selected on its ability to catch customers before they go silent.

Logistic Regression · CART · Feature Engineering

Healthcare Analytics × Power BI × Decision Support

Ready for Winter?

Pneumonia is predictable to the week and preventable by known means. The question is whether the health system is positioned before the peak arrives.

9 weeks

Mobilisation window before the regional influenza signal peaks in week 7.

Power BI · Python (pandas) · GBD / WHO data · Forecasting

Data Processing × Databases × Knowledge Graphs

Career Transition Data Pipeline

Skill demand is scattered across job boards, occupational standards and macro indicators. No single source tells a student what to learn or where to go.

93 of 133

Demanded skills matched to O*NET technology entries, 54 flagged Hot Technology.

MongoDB · Neo4j · RDF / SPARQL · Python

Agentic AI × NLP × Evidence Validation

Grounded Documentation Agent

ERP documentation fails in one direction: it asserts controls the configured system does not enforce. The agent has to refuse to write those claims.

19 rules

Deterministic validator between the model and the document. Claims resolve to evidence or go unpublished.

Python · State machine · TF-IDF retrieval · JSON Schema

Back to the homepage