Oussama B.

Technical referent within the team — 10 people (Java DevOps, Data Engineers, Data Scientists). Piloted the industrialization of 30 Kafka Streams applications and 30 Spark Scala reporters. Reduction of approximately 12% in Azure costs through governance and provisioning optimization.

Performance & JVM

• Advanced profiling (JVisualVM, JConsole, Eclipse MAT) — resolution of memory leaks, thread contention, GC pauses

• Java 8 → 17/21/22 migration and Spring modernization across all Kafka Streams pipelines

Data Streaming & Kafka

• Industrialization of 30 Kafka Streams applications and 30 Spark Scala reporters

• Kafka Connectors, MirrorMaker 2 (inter-cluster replication), Schema Registry (Avro)

• Kafka cluster supervision on Kubernetes (OpenShift, Azure) — Prometheus/Grafana/ELK monitoring

• Benchmarking Spark Streaming vs Kafka Streams, tuning retention/compression/consumer lag

AI Engineering

• Multi-agent RAG system integrated with Prometheus → Spark job diagnosis reduced from 4h to 15 min

• AgentOps Platform: 6 specialized AI agents (code review, security, monitoring, incident, fix + GitHub PR) via supervisor with dynamic re-planning

• Model Context Protocol (MCP) integration — 5 tool servers, real-time SSE communication, event-driven architecture on Azure Service Bus

• Enterprise multimodal RAG: dual-content indexing, hybrid search BM25 + vector search, cross-encoder ms-marco, 3 modes (Fast/Advanced/Agentic with self-refinement and confidence gating) → 90%+ accuracy

• Azure Deployment: Container Apps, AKS with self-hosted embedding & reranking models, Azure AI Search, Azure OpenAI GPT-4o

• Multi-turn conversational memory: anaphoric reference detection, contextual rewriting

Stack: Java 8/17/21/22 · Spring · Kafka · Spark Scala · Azure · Databricks · Azure AI Search · Azure OpenAI · Kubernetes · OpenShift · Airflow · Prometheus · FastAPI · LangChain · LangGraph · MCP

Tech Lead for the critical ROR (Recovery of Operations & Irregularities) platform processing 1.5 to 2.5 million messages/day in real-time (delays, cancellations, rebooking). Team of 12 developers (7 ROR + 5 Ticketing).

Key results:

• -38% execution time and -30% resources on the irregularity processing batch

• -90% of production incidents requiring manual support intervention

• Mentoring 20 engineers (multi-team): workshops, technical documentation, reviews on Java multithreading and JVM tuning

Technical achievements:

• JVM Optimization: heap tuning, GC (G1 → ZGC), thread dumps, thread contention resolution

• Java 8 → 11/17/21 migration and Spring modernization on critical services

• Intelligent load balancer with adaptive throttling and real-time token/API monitoring

• Adaptive HikariCP manager with autoscaling on runtime metrics (P95)

• Apache Ignite distributed cache, Oracle → PostgreSQL migration

• Azure Migration — batch workloads on Kubernetes (AKS)

• CI/CD Jenkins/ArgoCD GitOps + Chaos Testing

Stack: Java 8/11/17/21 · Spring Boot · HikariCP · Apache Ignite · Redis · Kafka · PostgreSQL · Oracle · Kubernetes (AKS) · Azure · Jenkins · ArgoCD · Prometheus · Grafana · ELK

Joined the PRODUCE AND DELIVER team, composed of 4 sub-teams on the PRODCOM project within the ORDER domain, responsible for managing internal orders and data exchange between PRODCOM and other systems (SAP, TRANSPORT, FINANCE).

Missions:

Development: bug fixes and new features

Batch optimization: multithread development and profiling

Batch and flow management / server instance management / hprof analysis

Technical and functional analysis of incidents and issues

RUN duty (1 week out of 6)

Technical Stack: Java 11, Struts, JSP, Spring Boot, Docker, REST API, Kafka, MQ Flows (via OpenEX), JUnit, Jenkins, GitHub (Git), Log4j, Swagger, Oracle, Hawtio, Rundeck, GoldenEyes, Kubernetes.

Methodologies: SCRUM, TDD.