Karim W. ElAzzouni

Meta Platforms — Senior Software Engineer

London, UK

Monetization AI Foundations: Training Reliability (July 2025 – Present)

Joined as a founding engineer of Meta's newly formed Monetization AI Foundations group in London, focused on the AI training reliability of a fleet of ads recommendation systems models. The team's flagship model, Generative Ads Model (GEM), is the industry's largest and most sophisticated RecSys model, trained on several thousand SOTA GPUs. Runs the productionization, developer efficiency, and trunk reliability programs across a 100+ person organization — leveraging agentic workflow techniques to boost productivity and building frameworks that keep the fleet efficient, available, highly observable, and maintainable.

• Developed a testing framework that saved 5.5K engineering hours across the org.
• Developed a trunk release framework that dropped average trunk staleness from 28+ to under 5 days, and average/maximum incident time-to-mitigate by 40% and 70% respectively.
• Both frameworks directly captured and reverted several OSS regressions in PyTorch (#171884, #191024).
• Introduced several agentic workflows to proactively detect, diagnose, and fix failures automatically, driving a 70% SLO attainment improvement.
• Created and ran several oncall loops to uplevel the reliability of production workflows.

Core Ads Quality (December 2021 – July 2025)

Part of the central team owning the organic bid component in Meta's ads auction system — building cross-app and infrastructure ads delivery solutions that optimize for long-term gain. As single-threaded owner, lead, or sole contributor across key workstreams, delivered:

• 0.17% increase in ads revenue per annum
• 2.5% potential ads revenue unlock through personalization
• 8.892 Engineering Years Saved (EYS) across developer productivity projects over five years

Long-term Ads Engagement (January – March 2022)

Built the first version of a complex data pipeline using Meta's internal Apache Airflow-based "Dataswarm" system to ingest ad engagement and quality metrics from multiple sources, train pre-existing ML models, and generate features and scalars for an experimental bid component aimed at improving long-term ad engagement.

Ads Conversion Familiarity "ACF" (March 2022 – September 2023)

Engineering owner of a new hypothesis to increase long-term revenue and user engagement by boosting familiar, high-quality ads — helping users convert on purchase ads and become recurring purchasers across Meta products.

• Started as the sole engineer; built and launched tens of A/B tests from ideation through shipping.
• Built the first delivery infrastructure for the initiative from scratch.
• Onboarded 6 engineers and grew the workstream to 20+ people across 3 time zones, evolving it into a stand-alone pod.
• Launched the new treatment to production, generating a 0.17% increase in ads revenue per annum.
• Received strong leadership recognition for impact on Meta's ad ecosystem.

Developer Productivity and Foundations Pod (September 2023 – July 2025)

Co-founded the Developer Productivity and Foundation pod to promote efficiency and reliability across infrastructure — owning projects E2E from ideation through adoption.

• UQL Testing Framework "UTF": Built an integration testing framework from scratch, integrated with CI and Testing Infrastructure, and drove adoption across the team and partner teams. Saved 1.8 EYS and prevented dozens of bugs and revenue-impacting SEVs from reaching production.
• User Features Ingestion Platform: Built a dedicated platform enabling engineers to define and consume user features within the ads delivery system, unblocking personalization efforts for ranking alterations. Potential impact of a 2.5% increase in ads revenue per annum. Saved 2.71 EYS.
• Experiment Monitoring, Management and Reporting Platform: Automated experiment metric collection from multiple data sources and auto-generated standardized Google Docs reports with scheduled email delivery to stakeholders. Saved 2.7 EYS, won "Best Time-Saving Hack" at the organization's Tooling Hackathon, and received Director recognition.

Microsoft — Software Engineer II / Software Engineer

Copenhagen, Denmark

Software Engineer at Dynamics 365 Finance and Operations in the Resilient Sustainable Supply Chain (RSSC) team, working across several platform services that other application teams across the product build upon.

Batch Processing Service

Became the official maintainer and subject matter expert of the Batch Processing Service — a light-weight scheduler and execution engine for asynchronous jobs across the Finance and Operations product, built in C# and X++. The service supported multiple flavors of jobs including application-team jobs, customer-authored jobs in custom deployable packages, and system jobs.

• Designed and rolled out Batch Priority-Based Scheduling (PBS) — a feature that required effectively rewriting the scheduling and polling flows in the service. PBS enabled job prioritization, completely decoupled batch from the underlying infrastructure, and unlocked two long-anticipated platform capabilities: elastic compute (dynamically adding/removing batch instances on-the-fly) and zero-downtime servicing (the product's approach to blue-green deployments with near-zero disruption).
• Managed backlogs, customer escalations, and acted as the central point of contact across customers, PMs, EMs, support engineers, and developers from different teams.
• Authored dozens of monitoring rules in Watchdog (an in-house rule-based production monitoring and detection tool) for proactive monitoring of the batch service.
• Contributed to Geneva Workflows and Actions — a low-code automation engine for production operations and incident response.

Trials Deployment Pipeline (AxTrials2)

Led the design and development of a new deployment pipeline for Finance and Operations trial environments — a modernization overhaul of the legacy trial system, built on top of in-house deployment services and Service Fabric clusters.

Cloud and Edge Validation Service

Designed, implemented, and owned the Cloud and Edge Validation Service — an internal distributed integration testing framework enabling application teams to write and execute distributed test scenarios against environments with cloud and edge capability — a new feature allowing customers to work on portions of the application offline.

Other Platform Contributions

• Member of the Cloud and Edge Platform team, building platform capabilities to extend the application to support distributed workloads.
• Contributed across feature management and exposure, number sequences framework, database synchronization engine, and the application's legacy kernel.
• Responsible for transitioning the batch service and platform knowledge to new owners through brownbag sessions, Troubleshooting Guides (TSGs), and comprehensive documentation.

Careem — Software Engineer

Dubai, UAE

Software Engineer in the Captain Payments team, responsible for the end-to-end process of payments to captains (Careem's term for drivers), pooling trips verification, and cash trips blocking/unblocking.

Captain Earning Service

Co-built the Captain Earning Service from scratch — an internal web service acting as the single, reliable source of truth for all captains' transactions.

• Decreased querying time by 80+% compared to the previous system.
• Introduced real-time captain balance calculation and real-time cash trips blocking/unblocking — a feature anticipated since Careem's inception.
• The service listened to events from the booking flow (monolith "core" service), computed payment splits across cash, credit card, and loyalty points, maintained debt tracking tables, and triggered block/unblock flows.
• Integrated the Captain Earning Service with Careem Solutions (internal customer support portal), migrating all direct database access flows to go through the new service via feature toggles.

Other Contributions

• Main contributor to Careem's most critical microservice, core-service — the former monolith containing booking, pooling, routing, and other essential flows.
• Maintained the cash blocking and pooling routes verification scripts.
• Built a QA bot as a serverless AWS Lambda function to automate clean test data creation, simulating external transactions through pay-out vendors (e.g., Fawry in Egypt, Easypaisa in Pakistan).
• Built the first version of an API backend for the Marketplace data science team to control their ML pipeline parameters.
• Onboarded multiple new team members and cross-team contributors.