Software Engineer II, Microsoft Cairo, Egypt Open to relocation

Bavly Shehata

I build the high-scale backends behind everyday products, and ship the machine learning that makes them pay off.

I lead the service behind affiliate links across Outlook, Microsoft Edge, and Copilot: a high-scale backend that handles hundreds of millions of requests a day and generates tens of millions in annual revenue, built hands-on with a small team.

§ 01 Selected work
01

Flagship · Machine learning · Chromium (C++)

On-device ML inside Microsoft Edge

Initiated and led replacing Edge's rule-based product-page detector with a single on-device machine-learning model, in the Chromium source.

  • Improved detection accuracy by ~14% and drove a 10% increase in daily ad calls by triggering Edge shopping features more accurately.
  • A tiny on-device model (a few kilobytes) with a single forward pass: no per-locale rules, no live API dependency, no page scraping.
  • Built the offline evaluation harness (headless tracing plus capture) and shipped the C++ serving-path gate, unit-tested and regression-clean.
Logistic regressionFeature engineeringModel validationC++Chromium
02

Ownership · Distributed systems · .NET

A monetization service at scale

I lead and build the service behind affiliate links across Outlook, Microsoft Edge, and Copilot.

  • A high-scale backend that handles hundreds of millions of requests a day and generates tens of millions in annual revenue.
  • Integrated multiple affiliate and retail networks (including a direct eBay integration), with yield optimization that routes each click to the highest-paying network.
  • Low-latency merchant resolution (exact, then base, then wildcard domain, ccTLD-aware) with signed redirect URLs.
  • Set direction and delivery with a small team, while staying a primary builder.
.NETC#MicroservicesAzureA/B experimentation
03

Data pipelines · Reliability

A revenue data platform teams can trust

Designed and refactored the pipeline and dashboard that make partner revenue accurate and fast.

  • Normalizes currencies and time zones, dedupes transactions across networks, computes commissions, and joins click data in memory.
  • Reliable alerting and a focused stakeholder dashboard that loads near-instantly.
  • Canonical retry and resilience patterns across SQL, Kusto, and HTTP, no bespoke retry logic.
SQLKusto (KQL)Data normalizationResilience
04

Reliability · Root-cause

Finding the expensive problems

Sometimes the fix is technical. Sometimes it is challenging the report everyone already trusts.

  • Recovered significant annual revenue by root-causing a production cluster outage, a partner had quietly blocked our traffic.
  • Broke a stalled, multi-week, all-hands incident by re-framing a mislabeled root cause, growing partner traffic ~10%.
  • Improved the reliability of Microsoft Cashback by ~20%.
Log analysisIncident leadershipCross-teamObservability
05

Developer productivity

AI-assisted development, for the whole team

Pioneered agentic coding workflows so the team ships faster without lowering the bar.

  • Wired GitHub Copilot and Claude Code into testing, documentation, code review, and refactoring.
  • Cut cycle time and helped set an AI-first engineering culture, plus a company security initiative (governed release pipeline, hardened storage).
GitHub CopilotClaude CodeAgentic toolingAutomation
§ 02 About

I am a software engineer who likes the impactful end of the work: taking a feature from an idea to something running in production, with all the cross-functional messiness that involves. My home base is backend and distributed systems, but I follow the impact wherever it leads, which is how I ended up inside the Microsoft Edge source, where I also lead the on-device machine-learning classifier that powers its shopping features.

None of this was the plan at 15, when an introductory C++ course turned into a love of breaking problems into logical steps (a first-place robotics win did not hurt). That led to a full-ride scholarship, a 4.0 Computer Science degree, years of competitive programming, and a long stretch of self-employed full-stack work before Microsoft.

I care about how we build, not just what we build: clean, well-tested code, strong fundamentals, and AI-assisted development. I also enjoy the human side, mentoring engineers and presenting the high-impact work I get to be part of.

Now

Leading the affiliate service, going deeper on AI-powered developer productivity and large-scale systems, and learning Spanish.

ChessOnline-privacy advocateLinux (EndeavourOS) tinkererLearning Spanish
§ 03 Capabilities

Backend & distributed systems

.NET / C# · Microservices · REST APIs · SQL · Azure · Kusto (KQL) · Data pipelines · A/B experimentation · Resilience & retry · Observability

Machine learning

Logistic regression · Feature engineering · Model validation · Model compression · PyTorch · Offline evaluation

AI-assisted development

GitHub Copilot · Claude Code · Agentic LLM tooling · Developer productivity

Full-stack & languages

Node.js / Express · React · HTML / CSS / JS · Python · C++ · Java · Git

§ 04 Background
2024 — present

Software Engineer II / Microsoft, Bing Shopping

Lead the affiliate service across Outlook, Edge, and Copilot, and the on-device machine-learning classifier that powers its shopping features. Recovered significant annual revenue and championed AI-assisted development. Promoted to Software Engineer II in 2026.

2021 — 2024

Full-stack Developer, Self-employed / Bavlifyweb

Delivered full-stack and low-code web projects for 20+ clients end to end, including FL0, PowerHouse SMART, and Trainline. Mentored 7 interns, built real-time integrations and automations, deployed on cloud.

2023

BSc Computer Science (AI major) / British University in Egypt

4.0 GPA, distinction with honors, merit-based full-ride scholarship. President of the university's ICPC (competitive programming) team. Degree project: multi-classification from facial images, 400+ models over 1,000+ GPU-hours (F1 73% age, 80% gender, 93% ethnicity).

Earlier recognition
  • US Embassy exchange scholarship to the Education USA Academy (University of Tennessee), one of about ten Egyptians selected.
  • Youngest of five students chosen to speak at a national STEM conference, covered in the national press.
  • NASA Space Apps Challenge: the only team from my school selected to travel and present.
  • First place in a local robotics competition, plus an ISEF science-fair project.
§ 05 Contact

Up for a good engineering problem, a role, or a game of chess. Let's talk.

bavly@bavlifyweb.com linkedin.com/in/bavshehata github.com/bavshehata