Mohamed Saadi
about / work / cases / arch

// Case Studies

Three Systems. Three Contexts.
Real Constraints.

Detailed engineering case studies — context, constraints, architecture decisions, what almost failed, and results with denominators. Not corporate storytelling. Engineering reasoning.

  • 2019–2021 · Enterprise MLOps

    AI/ML Platform at EDF

    Building a production AI/ML platform from scratch for 10+ data science teams. 100+ daily training jobs. 9× faster deployment. 30% cost reduction.

    9× Time-to-prod
    30% Cost reduction
    99.9% Uptime SLA
    KubernetesAirflowMLflowKafkaFastAPIAWS
  • Side Project · Agentic AI Architecture

    OrchestraAI

    Production-grade multi-agent ML platform with human-in-the-loop. Explicit state machine orchestration. 4 domain-expert agents. FastAPI + React.

    4 Agents
    3 HITL checkpoints
    HITL First-class state
    LangGraphFastAPIReactWebSocketDocker
  • EDF · Data Drift Remediation

    PrevCtGazTr — Gas Forecasting

    Diagnosing and remediating systematic data drift in a daily gas forecasting pipeline. 847 industrial PCEs. MAPE reduced from 2.44% to 1.83% — a 25% improvement.

    847 PCEs
    1.83% Final MAPE
    −25% vs baseline
    LightGBMOptunaPanderaPythonPandas
© 2025 | Mohamed Saadi