Specialist track

AI / LLM Engineering

The differentiator track for working engineers. Go beyond calling an API: design retrieval systems, evaluate model output, control cost and latency, guard against failure, and ship AI features your users can trust. Part-time, project-driven, taught by practitioners.

Duration: 12 weeks
Commitment: ~12 hrs/week · part-time, evenings
Format: Live online
Level: Intermediate · for people who already write code

Specialist track

$350

Indicative · your local currency · pre-VAT-determination

Deposit $90, then monthly installments

Cohort 1 · Online

Starts 28 September 2026

Filling fast

Apply now

A short application. No payment on this site.

Refund & deferral policy

What you'll learn

Prompt & context engineering, structured output, and tool/function calling
Retrieval-augmented generation: embeddings, vector stores, chunking that works
Agent loops and graph-based agent orchestration. Multi-step workflows with state, checkpointing and recovery
Connecting agents to tools and data with MCP
Evaluation: how to actually measure whether your AI feature is good
Cost, latency, safety and guardrails for real users

AI-native by default

The AI classes in this program

Every Lightspace program teaches the AI-native method. Not a bolt-on module, but a working style woven through the projects.

Context engineering as a first-class discipline. The skill that replaced prompt engineering
Agent loops and harness engineering. Designing the constraints an agent works inside
Graph-based agent orchestration (LangGraph): state machines, checkpointing, recovery
Evals. Systematic measurement of AI output quality, not gut feel

The curriculum

1
LLM & context foundations
Weeks 1, 3
How these models actually behave, prompt and context engineering, structured output, tool calling, and building your first reliable pipeline.
2
Retrieval & data
Weeks 4, 6
Embeddings, chunking, vector search and RAG patterns that survive contact with real documents and real users.
3
Agents: loops, graphs & tools
Weeks 7, 9
The agent loop, harness engineering, and graph-based orchestration with LangGraph. Multi-step agent workflows with state, checkpointing and recovery, connected to tools and data via MCP.
4
Evals, reliability & ship
Weeks 10, 12
Building evals, measuring quality, cost and latency control, guardrails, and a capstone agentic feature you present and defend.

What you leave with

A deployed agentic LLM feature with retrieval, evals and guardrails
A repeatable evaluation harness you can take to your job
The vocabulary and judgement to lead AI work on your team

How you're taught

Live cohorts led by working practitioners. Not pre-recorded videos.

Live, not recorded

Real classes on a schedule, with a group that keeps you moving.

Practitioner instructors

Taught by people who build and ship software for a living.

Feedback on your work

Code review and honest feedback on real projects, every week.

Questions

Where does it take place?

Live online. You can join from anywhere in the world.

How much time does it take?

12 weeks, at roughly ~12 hrs/week · part-time, evenings.

What do I need to start?

Intermediate · for people who already write code.

How do payments work?

There's no payment on this website. You submit a short application; if you're accepted, our team emails you to confirm your place and arrange payment. A deposit, then monthly installments.

What language is it taught in?

All live classes and materials are in English.

Is this for beginners?

No. It's for people who already write code. If you're newer, start with a short course or the AI-Native Software Engineering bootcamp.

Full details in our Refund, withdrawal & deferral policy.