Senior Data Scientist

We are a fintech startup building a merchant cash advance product for Sri Lankan SMEs, inside one of the country’s largest payment networks, with proprietary access to building blocks others cannot easily replicate. The product is live, merchants onboard every week, and new banks and collection rails are coming online - we are well funded and focused on speed and scale.

**THE ROLE**

**This is a hands-on role for someone who wants to own a risk model from first advance to first retrain.** We advance funds against a merchant’s daily card settlements and collect through a variety of payment methods. We decide who qualifies, and for how much, from transaction history and other data sources. Today that runs on transparent, documented rules. You run them as the book grows, turn every advance into training data, train the first real models once there are enough outcomes - and prove they beat the rules.

We are building for the AI-led era - a team that runs at high capacity rather than one that scales by headcount. Today that is the CEO, a Head of Operations and an operations team, a Technical Product Manager (hiring now), a contract engineering team, and a data science advisor who built and documented the scoring approach. You take it on from day one - you are the data function.

**WHAT YOU'LL OWN**

• **The decision** - the scoring and offer-sizing rules behind every advance: who, how much, and why. Transparent, versioned, defended on the numbers
• **The path from rules to models** - labels, decision-time snapshots, the first trained scores; shadow, champion-challenger, promotion, with the rules as fallback
• **How the book is doing** - collection pace, early warnings, shortfall and recovery by cohort, so the team acts on numbers
• **The merchant data** - bank and network extracts reconciled, de-duplicated and versioned; you know the gaps and chase the data owners until they close
• **The numbers the business runs on** - eligible pools, cohort sizing, expected loss, seasonality - ready when the CEO and ops need them

**WHAT YOU'LL BUILD**

• **The label and feature pipeline** - merchant-day series, decision-time snapshots and advance outcomes, versioned by data vintage, so the first model trains leakage-free the week labels exist
• **The first trained scores** - a behaviour score for collections, an application score for new merchants, and the harness that decides whether they replace the rules
• **The reporting the team runs the book on** - collections views, cohort read-outs, outcome tables by band. Scripts and dashboards, not slide decks

**WHO YOU ARE**

• 3-5+ years of applied machine learning in production with real decisions riding on it - you’ve owned a model end to end and know when a percentile rank beats a deep net
• Fluent in Python (pandas, scikit-learn, gradient boosting) and SQL; you build with AI tooling - Claude Code, Cursor and the like - and ship real work with it
• You turn raw logs into clean features, defend every number, and design pilots that yield next year’s labels
• Risk scoring, payments or SME working-capital experience is a strong plus - scorecards, PD models, collections, reject inference - or show us you learn new domains fast
• A doer. You chase the missing extract, close the loop with the data owner, and take pride in the unglamorous work that makes a model trustworthy.

**WHY THIS ROLE**

• A rare starting point - a live book and a documented scoring approach; you build the first trained model on top
• You are the data function - a direct line to the founders and your call on the detail

• Backed by Sri Lanka’s largest independent payment network - distribution and data are already there