QueesAILearn any skill from scratch
QueesAI teaches through short, hands-on labs. Each one starts at zero, explains every new word in plain language, and lets you try it yourself right away. No setup, no account.
Reach a goal: follow a program
A program is a short path through several labs toward one goal, with a tick-list so you know where you are.
Build an LLM from scratch
Train a small language model on your own device and watch it learn to write short stories.
Open the programFind the right information: retrieval from scratch
Learn how search finds the right page for a question, one technique at a time, so a small AI can answer from documents it has never seen.
Open the programTrain a model through its life cycle
Take one tiny model through the stages that turn raw text into a helpful assistant, and see what each stage gives and what it costs.
Open the programLabs
Or explore one topic. Each lab assumes you are new to the topic.
AI
Run, build and train AI models yourself, right in your browser.
How AI runs on your own device
Download a small AI model into your browser and chat with it, then build a talking assistant that listens and answers out loud. A step-by-step guide and a word list walk you through everything.
Open the playgroundHow transformers work, from 2017 to today
Build a tiny AI model of the kind behind chatbots, train it live in your browser, and switch on each big idea since 2017 to see exactly what it changed. Every new word is explained.
Open the transformer labBuild your own tiny LLM
Download thousands of short children’s stories, cut them into tokens, and train a small language model from zero on your own device. Then ask it to write a story. Every new word is explained.
Open the LLM labGive an AI your own documents (RAG)
An AI only knows what it read in training. Here you hand it a few pages about a made-up island town, and try four ways to find the right page for each question: plain search, rewriting the question, following a map of names, and passing whole documents. Every score is shown.
Open the RAG labKeyword search from scratch
Build the kind of search that finds the right page by its words: an index, yes/no Boolean search, then TF-IDF and BM25 scores. You see every number behind each ranking, and move the sliders yourself.
Open the keyword search labFuzzy search: find it despite typos
Exact search misses “Okafur” when the page says “Okafor”. Try four ways to forgive mistakes: edit distance, shared letter pieces, sound-alike codes and autocomplete. You fill in the tables and check every number by hand.
Open the fuzzy search labVector search: find by meaning
Turn text into lists of numbers that capture its meaning, then find the closest ones. Compare exact search with faster approximate search, keep the vectors in your browser, and see a question find the ferry fare without sharing a single word with it.
Open the vector search labHybrid search and reranking
Run keyword and vector search together, merge their lists with rank fusion, filter by label, let a reranker take a second look, and see on 36 questions which method wins where. None of them wins everywhere.
Open the hybrid search labPut it together: answer and measure
Pick a search method, see the passages and the prompt, ask a small model to answer, then score every method with recall@k and MRR. Open any number to see the sum behind it, and learn which technique to use when.
Open the put-it-together labMid-training: teach a model a new kind of text
Take a tiny story-writing model and keep training it on a mix of stories and word sums. Choose how much is new, run it for a few minutes in your browser, and see what it gains (answers that look like sums) and what it forgets (how to write stories). Every number is shown.
Open the mid-training labSupervised fine-tuning: teach a model to follow an instruction
Show a tiny story model examples of a prompt (“name: Bob”) and the answer you want (a story about Bob). See which tokens count towards the loss, choose how many examples to train on, and see what it learns and what it costs. Every number is shown.
Open the fine-tuning labHuman feedback and RLHF: teach a model what people prefer
Be the human: say which of two stories is better, and see how long it takes. Train a reward model on thousands of labels, then let the story model learn from its scores, with a leash that stops it cheating. Then try your own taste. Every number is shown.
Open the human feedback labDPO: learn straight from preferences
Train the story model directly on pairs of better and worse stories, with no reward model and no new stories written while it trains. Compare it with RLHF on kindness, cost and time, try it on your own taste, and meet the methods that came after it. Every number is shown.
Open the DPO labMath
The math behind the ideas, worked through one real number at a time.
Physics
How the world moves, with experiments you can change and rerun.
Physics labs are on the way
They will follow the same format: learn the idea, try it yourself, see what happened. Want a topic covered? Tell us at contact@queesai.com.
How a lab works
Every lab follows the same three steps, so you always know what comes next.
Learn the idea
A short, plain-language explanation of what you are about to try. New words come with a definition.
Try it yourself
Follow guided first steps, then experiment freely. You can’t break anything.
See what happened
The lab shows what went on behind the scenes, so you understand why it worked.