Senior Data Scientist — Transaction Intelligence

Senior Data Scientist — Transaction Intelligence

Application Deadline: 31 August 2026

Department: Engineering

Employment Type: Full Time

Location: Riyadh

Description

About the role:

Malaa is Saudi Arabia’s first retail Open Banking platform in the Kingdom. Our product is built on one core asset: hundreds of millions of records of behavioral data. Turning that raw data into something people can act on is this role's whole job, and it spans the full pipeline: enriching transactions (categorizing short, noisy merchant strings in mixed English and latinized Arabic, and resolving them to real merchant entities), then building the models and analyses on top that turn transaction data into insight across our services — categorization, pattern recognition, anomaly detection, and risk assessment.

You will be the data scientist for transaction data across the company: owning the enrichment models end to end (modeling, evaluation, and the production code), and partnering with product and engineering teams to design, ship, and measure model-backed features in other services.

What you'll work on

• Own and level up transaction enrichment — the foundation everything else stands on: categorization that generalizes to merchants we've never seen, and merchant name matching that tolerates harmless variation without confusing genuinely different businesses.
• Own model confidence across our systems: well-calibrated probabilities, principled abstention on uncertain cases, and confidence-based routing that products and review workflows rely on.
• Build transaction-based insight models for other services: pattern recognition, anomaly detection, and behavioral analysis on transaction streams.
• Own the ongoing validation of our risk models: backtesting scores against realized users behavior, monitoring discrimination and calibration as behaviors evolve, and driving model improvements from what the data shows.
• Handle our text and behavior data as they really are: bilingual, informally romanized Arabic with unstable spelling; transaction streams with truncation artifacts, bank quirks, and heavy-tailed distributions.
• Build efficient batch pipelines at hundreds-of-millions-record scale, designed to be re-run routinely as models improve.
• Design within the guardrails of a regulated fintech: data governance, privacy, and cybersecurity requirements shape what data we can use, where workloads can run, and which tools we can adopt — you'll build excellent solutions inside those boundaries, working with those teams rather than around them.
• Build the evaluation discipline these systems deserve: labeled datasets, regression test suites for model behavior, and metrics that answer "did this change help?" for every release.

Must-have qualifications

• 3+ years of applied ML/data science, with models shipped and maintained in production at scale — and knowledge of their failure modes.
• Strong analytical range beyond modeling: exploratory analysis, statistical rigor, and feature design on behavioral/tabular data (SQL fluency assumed).
• Hands‑on experience with text similarity, fuzzy matching, or entity resolution on noisy real‑world strings.
• Strong Python and the scientific stack (scikit‑learn, scipy, numpy), with performance and cost awareness at scale — vectorization, sparse data structures, efficient batch computation.
• Demonstrated maturity working under externally imposed constraints — data governance, privacy, cybersecurity, compliance, infrastructure policy — including collaborating with the teams who own them.
• A track record of turning ambiguous "the model feels wrong" complaints into measured, regression‑tested properties.
• Comfort reading and debugging model code you didn't write, and working directly with product teams on loosely‑defined problems.

Nice-to-haves

• Arabic / Arabizi text processing — a strong plus; our data is bilingual with unstable romanization.
• Credit or behavioral risk modeling and validation — scorecards, discrimination and calibration measurement, backtesting against realized outcomes.
• Text embeddings and approximate nearest‑neighbor retrieval in resource‑conscious settings.
• LLM‑assisted labeling, distillation, or weak‑supervision pipelines.
• ML lifecycle and batch‑serving tooling (experiment tracking, data versioning, distributed task queues).