About
I'm Firaz Zakariya, a data scientist and analyst based in Sweden. I'm interested in applied machine learning, the kind that lives close to real decisions rather than benchmark leaderboards.
I trained as a mechanical engineer (BTech, First Class) before moving into data science via an MSc in Computational Software Techniques in Engineering at Cranfield, where I graduated with Distinction. My thesis was on explainable machine learning for flight-delay classification: early grounding in models that are accurate enough to use and transparent enough to defend.
Since then I've spent four years building forecasting, classification, and scenario models for operations teams and senior leadership. Most recently at the Independent Office for Police Conduct (IOPC), I developed hybrid backlog forecasts, delay-risk classifiers with SHAP interpretability, and forward-looking scenario simulations under alternative resourcing assumptions. Work validated with out-of-time splits and adopted into capacity planning and process decisions. I also designed and analysed controlled A/B experiments with proper power analysis, not post-hoc significance hunting.
Alongside that, I'm co-founder of Cawosh, a subscription SaaS for UK independent garages. I own the metrics framework (recurring revenue, engagement, churn signals) and the product thinking around retention. That side project is why this site includes microexp: I like building analytics where the data pipeline and the business model are both things you have to trust.
The projects on this site cover experimentation, time-series forecasting, and analytics engineering. They reflect what I find worth building: systems where the output is interpretable and the data pipeline is something you can actually trust.
I'm open to data scientist and analyst roles in Sweden, especially where forecasting, risk analytics, experimentation, or analytics engineering matter: public sector operations, fintech and credit risk, telco provisioning, or product-led SaaS. I'm comfortable owning the full path from raw data to deployed model, and I'm used to presenting results to people who are not modellers.