Command-line assistant
A small script that answers questions through a model API, with your key kept in an environment variable.
Build useful, safe products on top of large language models and other AI models.
You have reached the end of the beginner stages. Prove what you learned with a project.
You have reached the end of the intermediate stages. Prove what you learned with a project.
You have reached the end of the advanced stages. Prove what you learned with a project.
A small script that answers questions through a model API, with your key kept in an environment variable.
Turn long job posts into a short list of duties, must-have skills and nice-to-haves.
Extract a CV into clean JSON and validate it, flagging anything the model was unsure of.
Answer questions from a folder of PDFs and show which passage each answer came from.
An agent that searches, reads and writes a short cited brief, with a step limit and a cost cap.
A test set and scoring script that runs on every prompt change and blocks changes that score worse.
The same roadmap as a reading list: every stage, what each topic covers, and which ones are optional.
The language most AI tools and examples use: functions, classes, packages and virtual environments.
How programs call each other over HTTP, send JSON and handle errors and rate limits.
Version your code and prompts, and collaborate through pull requests.
Run scripts, manage environment variables and keep API keys out of your code.
What a model is trained on, how it predicts text, and why it can sound sure while being wrong.
Models read tokens, not words. Tokens decide what fits in a request, how fast it is and what it costs.
Turning text into vectors so you can search and group it by meaning rather than exact words.
Hallucination, bias, outdated knowledge and sensitivity to wording, and how to design around them.
Send messages, set parameters, stream replies and handle timeouts and retries.
Clear instructions, examples, roles and step-by-step structure that make results reliable.
Get JSON your code can trust and let the model call your functions.
Run models on your own machine or servers when privacy, cost or offline use matter.
Answer from your own documents: split, embed, retrieve the right passages and cite them.
Store embeddings and find the nearest ones fast, often inside a database you already use.
Chain model calls and tools to finish multi-step tasks, and know when a simple workflow is enough.
Libraries that bundle retrieval, memory and tools. Learn the basics without them first.
Build test sets and scoring so you know a prompt or model change made things better, not worse.
Untrusted text can steer a model. Limit what it can do and never trust its output blindly.
Choose the smallest model that works, cache repeated context, batch jobs and stream answers.
Privacy, consent, fairness and being honest with users about what the AI does.
Wrap your model calls in an API with authentication, rate limits and clear errors.
Log inputs, outputs and ratings so you can spot drift and fix real failures.
Set expectations, show sources, allow corrections and keep a human in control.
Adapt a model to your style or task when prompting and retrieval are not enough.
They build products and features on top of AI models: choosing models, writing prompts, connecting data and tools, and measuring quality, safety and cost in production.
An ML engineer mostly trains and serves their own models. An AI engineer mostly builds on existing models through APIs or open weights. The skills overlap, and many people do both.
Not to start. Solid programming, clear thinking about data and careful testing matter more. Maths helps later if you move into training or fine-tuning models.
Rarely. It is one skill of an AI engineer, alongside software engineering, evaluation and product sense.
© 2026 KamerLink. All rights reserved.