Chapter 03: MLOps Roles & Team Structure🔗

"MLOps is as much about people and process as it is about technology."

3.1 The MLOps Team Landscape🔗

Successful MLOps requires collaboration across multiple roles. Here's how they fit together:

┌──────────────────────────────────────────────────────────────────────┐
│                     MLOPS TEAM STRUCTURE                             │
│                                                                      │
│  ┌──────────────────┐    ┌──────────────────┐    ┌───────────────┐  │
│  │  DATA SCIENTIST  │    │  ML ENGINEER     │    │ MLOPS ENGINEER│  │
│  │                  │    │                  │    │               │  │
│  │  Problem framing │◀──▶│ Model training   │◀──▶│ CI/CD pipelines│  │
│  │  Exploration     │    │ Optimization     │    │ Infrastructure │  │
│  │  Prototyping     │    │ Productionization│    │ Monitoring     │  │
│  └──────────────────┘    └──────────────────┘    └───────────────┘  │
│           │                       │                      │          │
│           ▼                       ▼                      ▼          │
│  ┌──────────────────┐    ┌──────────────────┐    ┌───────────────┐  │
│  │  DATA ENGINEER   │    │ PLATFORM ENGINEER│    │  QA ENGINEER  │  │
│  │                  │    │                  │    │               │  │
│  │  Data pipelines  │    │  K8s clusters    │    │  Data tests   │  │
│  │  Feature eng.    │    │  GPU infra       │    │  Model tests  │  │
│  │  Data quality    │    │  Developer tools │    │  Integration  │  │
│  └──────────────────┘    └──────────────────┘    └───────────────┘  │
└──────────────────────────────────────────────────────────────────────┘

3.2 Role Descriptions🔗

Data Scientist (DS)🔗

Focus: Building accurate models
- Explores data, hypothesizes features, trains models
- Uses: Jupyter, Python, sklearn, PyTorch, Pandas
- Produces: Model prototypes, experiment results, feature ideas
- MLOps responsibility: Write clean, modular code; use experiment tracking; document experiments

Machine Learning Engineer (MLE)🔗

Focus: Productionizing models
- Takes DS prototypes and makes them production-ready
- Uses: Python, Docker, FastAPI, MLflow, CI/CD tools
- Produces: Packaged models, serving code, tests
- MLOps responsibility: Refactor notebook code into modules; write unit tests; optimize inference

MLOps Engineer🔗

Focus: Automating and operating the ML platform
- Builds CI/CD pipelines, manages infrastructure, sets up monitoring
- Uses: Jenkins, Kubernetes, Terraform, Prometheus, Airflow
- Produces: Pipelines, dashboards, alerting systems
- MLOps responsibility: Everything runs reliably at scale

Data Engineer (DE)🔗

Focus: Reliable data pipelines
- Builds ETL/ELT pipelines that feed ML training and serving
- Uses: Airflow, Spark, dbt, BigQuery, Kafka
- Produces: Clean, versioned datasets; feature pipelines
- MLOps responsibility: Ensure data quality and freshness

Platform / Infrastructure Engineer🔗

Focus: Developer platforms and compute
- Manages Kubernetes clusters, GPU nodes, storage
- Uses: Terraform, GKE, Helm, Ansible
- Produces: Scalable, reliable compute infrastructure

3.3 Team Topologies🔗

Model 1: Embedded ML Teams🔗

Product Team A:         Product Team B:
  DS + MLE + DE            DS + MLE + DE
  (own full stack)          (own full stack)

Shared Platform:
  MLOps Eng + Platform Eng (serve both teams)

Best for: Large orgs with many independent products

Model 2: Centralized ML Platform🔗

                    Central ML Platform Team
                   (MLOps + Platform + DataEng)
                          │         │
            ┌─────────────┘         └─────────────┐
     DS Team A                              DS Team B
  (uses shared platform)               (uses shared platform)

Best for: Orgs starting MLOps, standardization needed

Model 3: Full-Stack ML Teams🔗

Team:  DS + MLE + DataEng + MLOps (all roles in one team)

Best for: Startups, early-stage ML

3.4 Responsibilities Matrix (RACI)🔗

Activity	Data Scientist	ML Engineer	MLOps Engineer	Data Engineer
Define business problem	R/A	C	I	I
Data exploration	R/A	C	I	C
Feature engineering	R	A	I	C
Model training code	R	A	I	I
Unit tests for model	C	R/A	I	I
CI/CD pipeline	I	C	R/A	I
Docker packaging	C	R	A	I
K8s deployment	I	C	R/A	I
Monitoring setup	I	C	R/A	I
Data pipeline	I	I	C	R/A
Feature store mgmt	C	C	I	R/A

R=Responsible, A=Accountable, C=Consulted, I=Informed

3.5 Communication & Collaboration Tools🔗

┌──────────────────────────────────────────────────┐
│           MLOPS COLLABORATION STACK              │
│                                                  │
│  💬 Communication: Slack, Teams                  │
│  📋 Project Mgmt:  Jira, Linear, GitHub Issues   │
│  📚 Documentation: Confluence, Notion, GitHub Wiki│
│  🔬 Experiments:   MLflow, W&B (shared access)   │
│  📊 Dashboards:    Grafana, Looker (shared)      │
│  💻 Code Review:   GitHub PRs with required      │
│                    reviews from MLE + MLOps      │
└──────────────────────────────────────────────────┘

Next → Chapter 04: Git & GitHub