Become an ML engineer: training, serving and monitoring models in production.

What ML engineers own, and how LLMs changed it.

1. ML vs AI vs data engineering
2. The production ML lifecycle
3. Classic ML and LLM systems
4. Skills employers ask for
5. Building a portfolio

Reproducible, scalable model training.

1. Data loading at scale
2. Training loops in PyTorch
3. Experiment tracking
4. Distributed training basics
5. Checkpointing

Make the most of expensive hardware.

1. GPU architecture basics
2. Memory: weights, activations and KV cache
3. Mixed precision
4. Profiling bottlenecks
5. Choosing GPUs

Adapt open models efficiently.

1. When to fine-tune
2. LoRA and QLoRA
3. Dataset curation
4. Preference tuning with DPO
5. Evaluating fine-tuned models

Serve models fast and cheaply.

1. vLLM and SGLang
2. Continuous batching and PagedAttention
3. Quantisation
4. Speculative decoding
5. Latency metrics: TTFT and throughput

Ship models like software.

1. Model registries
2. Automated validation gates
3. Canary and shadow deployments
4. Rollbacks
5. Infrastructure as code

Run ML workloads on modern infrastructure.

1. Docker for ML
2. Kubernetes essentials
3. GPU scheduling
4. Autoscaling inference
5. Ray for distributed work

Know when a model degrades.

1. Data and prediction drift
2. Latency and error monitoring
3. LLM tracing
4. Alerting
5. Feedback loops

Quality management for LLM systems.

1. Eval harnesses
2. RAG evaluation
3. Prompt and model versioning
4. A/B testing models
5. Cost tracking

Fine-tune, serve and monitor a model end to end.

1. Fine-tune an open model
2. Serve with vLLM
3. Deploy on Kubernetes
4. Monitoring dashboard
5. Write-up and demo