Modern ads platforms run on always-on, real-time data: streaming events, feature computation, near-real-time aggregations, and low-latency serving to power ML models that operate at massive scale under strict freshness, cost, and reliability requirements.

Microsoft Ads builds and operates large-scale, latency-sensitive systems that serve billions of requests. We are looking for a Principal Software Engineer who is hands-on with production coding and system design to build the real-time data pipelines and feature/embedding materialization systems that feed online stores/caches and integrate tightly with ML inference serving.

This role is ideal for engineers who enjoy:

• building robust streaming + ETL systems (correctness, idempotency, backfills, late data),
• owning SLOs with strong observability and operational maturity,
• and optimizing end-to-end performance and cost across compute, storage, and serving integrations.

Primary success metrics are freshness, correctness, latency, reliability, and cost in production.

• Design and implement real-time streaming ETL / feature pipelines (e.g., Flink or Spark Structured Streaming) that meet strict freshness and correctness constraints.
• Build and operate reliable messaging and ingestion with Kafka/Pulsar (partitioning strategy, retries, ordering guarantees, DLQs, backpressure handling).
• Own data contracts between producers, pipelines, and consumers: schema evolution, versioning, compatibility, validation, and safe rollout.
• Implement production-grade backfill/replay workflows
• Define and meet SLOs using OpenTelemetry/Prometheus/Grafana for metrics, tracing, dashboards, alerting, and incident response readiness.
• Integrate pipelines with online stores/caches and ML consumers (feature stores, embedding pipelines, LLM API calls, online/offline consistency patterns).
• Partner with applied scientists on feature/embedding definitions, validation, and end-to-end quality measurement.
• Optimize end-to-end performance and efficiency: CPU/memory/I/O, serialization, caching, network overhead, concurrency, and pipeline compute cost.
• Contribute to serving/inference integrations where needed (e.g., Triton/ONNX Runtime/TensorRT) including batching and latency/cost tradeoffs.
• Ship safely with CI/CD, automated testing (unit/integration/data quality), and operational playbooks/runbooks.

Required Qualification:

• Bachelor’s or Master’s degree in Computer Science, Electrical/Computer Engineering, or a related field, with 8+ years of related experience.
• Strong programming skills in language C++,C# or Python (at least one required).
• Hands-on experience in one or more:
  • Building and operating streaming data pipelines in production (Flink or Spark Structured Streaming),
  • Distributed systems engineering with strong reliability and operational rigor,
  • Messaging systems such as Kafka/Pulsar.
• Experience operating services with Kubernetes/containers and production readiness practices (deployments, scaling, rollbacks).
• Experience with observability stacks such as OpenTelemetry, Prometheus, Grafana.
• Ability to debug complex production issues using logs/metrics/traces and performance profiling.
• Strong communication and collaboration skills, with experience working across engineering, applied science/ML, and product/business stakeholders.

Preferred Qualifications:

• Experience with feature stores, embedding pipelines, and online/offline consistency (freshness guarantees, correctness validation).
• Experience with data lakehouse/table formats and optimizations eg partitioning, compaction, and incremental processing.
• Experience with GPU inference serving (Triton, ONNX Runtime/TensorRT) and performance techniques (batching, request shaping, tail-latency reduction).
• ​understanding of pipeline correctness patterns: idempotency, dedup, watermarking, late data, exactly-once vs at-least-once tradeoffs.
• Background in cost/performance modeling, capacity planning, and reliability improvements for high-scale data platforms.
• Experience in Ads/search/recommendations or other high-scale systems where freshness, latency, and cost are jointly optimized.

This position will be open for a minimum of 5 days, with applications accepted on an ongoing basis until the position is filled.