LOCKEDIN / FORWARD-DEPLOYED ENGINEERING 101
Understand it.
Then build it.
You can be an experienced engineer and still be new to agents. Start with what each thing is and how it helps a person do useful work. Explore 58 original introductions across software, agents, cloud, security and delivery. Start here without a model key; each topic connects to deeper instruction and practice.
A place to start.
Choose a short reading route, or look up a concept below.
- 101UnderstandPlain language, visual walkthrough, vocabulary
- 102Build with guidanceOriginal lessons, examples and knowledge checks
- 103Practise and explainRunnable labs, failure cases and handover
Understand agents
Start with what produces the text. Separate the model from the application using it.
See the 8-concept route
For engineers who know software but are new to AI applications. Open the related definitions whenever a term is unfamiliar.
- 01The modelYou are here
- 02TokensSee the units a model reads and produces before reasoning about its input limits.
- 03ContextDecide what information reaches the model for this particular request.
- 04The agent loopPut the model inside an application that can execute permitted actions and check the result.
- 05WorkflowsSeparate a fixed sequence from a model-selected next action before choosing a framework.
- 06LangChainSee how a framework assembles the model, tools and application-controlled context.
- 07LangGraphMake the steps and shared state explicit when work needs to branch, wait or resume.
- 08MCPConnect capabilities through a shared protocol while keeping permission checks in the application and server.
Bookmark this page to return to this concept. The number shows your place in the reading route; it is not an assessment result.
A trained statistical model
What is a large language model?
A large language model (LLM) is a trained neural network. Its learned numerical parameters and current input are used to calculate possible next tokens; a decoding process selects the continuation. A model can produce text, code or structured proposals. The application around it supplies context, decides what tools are available and controls real-world actions.
Choosing a model is one decision inside an application design, not the whole design.
UNDERSTAND THE SYSTEM
What changes when a model writes the next word?
Follow an inspectable word-bigram model. Training counts adjacent words; inference uses the learned table. This original teaching model is not an LLM: it sees only one prior word and has no neural network or attention. The downloadable program performs these calculations locally.
Read the connections in plain English
- 01Training examples → Learned table
The learning procedure creates the model's parameters from examples.
- 02Learned table → Next-token probabilities
Inference reads the saved values without updating them in this exercise.
- 03Request context → Next-token probabilities
The toy selects the row for the last word; a Transformer computes representations from its permitted context.
- 04Next-token probabilities → Token selection
The decoding strategy operates on the computed distribution.
- 05Token selection → Generated text
A selected word joins the working context for the next step; the end token or limit stops generation.
Check your understanding: If the output changes, must the model have been retrained?
No. Context, decoding settings and sampled choices can change an output while parameters remain fixed. Inspect the actual model/version, request, settings and result. A low temperature is not a correctness check.
CONNECT IT TO WHAT YOU KNOW
Start with something familiar.
Think of a powerful, learned text transformation function with uncertain outputs. Unlike a normal lookup, it can compose an answer that was never stored as a record.
Plausible output is not a verified fact. Training knowledge, information supplied in the current request and live tool observations are different sources.
Words worth knowing
- Inference
- Using a trained model to produce output.
- Training
- Learning model parameters from data.
- Hallucination
- Generated content that appears plausible but is unsupported or incorrect.
ONE EXAMPLE / FICTIONAL CASE
Follow the connections.
The model writes a convincing maintenance instruction, but the approved procedure may have changed since the information it learned.
UNDERSTAND THE SYSTEM
What changes when a model writes the next word?
Follow an inspectable word-bigram model. Training counts adjacent words; inference uses the learned table. This original teaching model is not an LLM: it sees only one prior word and has no neural network or attention. The downloadable program performs these calculations locally.
See what moves and why.
Read the connections in plain English
- 01Training examples → Learned table
The learning procedure creates the model's parameters from examples.
- 02Learned table → Next-token probabilities
Inference reads the saved values without updating them in this exercise.
- 03Request context → Next-token probabilities
The toy selects the row for the last word; a Transformer computes representations from its permitted context.
- 04Next-token probabilities → Token selection
The decoding strategy operates on the computed distribution.
- 05Token selection → Generated text
A selected word joins the working context for the next step; the end token or limit stops generation.
Check your understanding: If the output changes, must the model have been retrained?
No. Context, decoding settings and sampled choices can change an output while parameters remain fixed. Inspect the actual model/version, request, settings and result. A low temperature is not a correctness check.
Work through it, step by step.
Supply input
Provide the task, relevant approved evidence and output constraints.
What you have now: A bounded request
Generate
The model produces tokens conditioned on that input. Different models and settings can produce different responses.
What you have now: A proposed answer
Check
Verify important claims against current sources and apply application rules before acting.
What you have now: A reviewed result
FROM RECOGNITION TO REASONING
Separate training, inference and token selection in an inspectable model.
- 01
Predict and run seven cases in the free Model 101 guide.
- 02
Change the sampling seed and extra training examples separately; compare parameters and outputs.
- 03
Explain which limits belong to the tiny word-bigram model and what a neural LLM does differently.
Seven traces, two controlled edits and a model/application diagram.
Compare your reasoning when you are ready
A different output need not mean different learned parameters. Context, decoding and sampling also matter. The count model demonstrates those roles without claiming to implement neural attention or factual understanding.
Keep notes in your own working copy. Moving through these cards does not record completion.
TURN THE IDEA INTO USEFUL WORK
You can now go deeper.
Plausible output is not a verified fact. Training knowledge, information supplied in the current request and live tool observations are different sources.
Academy lessons use your existing training access. Free guides stay open.
Learn with the original creators
A conceptual introduction published in 2023. Use it for the mental model; use current provider documentation for product setup.