Answers across several sources
Combine prose, scanned notices, and dated tables. Check whether retrieval covers every part of a question and choose the applicable source when facts appear to conflict.
Curriculum roadmap
See what you can learn today and the topics planned for later editions.
Available now
From first prompt to local project
61 lessons cover AI foundations, local runtimes, projects, and evaluation.
Curriculum updates
Offline voice, document editions, and model training
Build a voice guide to a PDF, search the right version of a table, compare vector indexes, and train a small Transformer on your CPU.
- Extract PDF tables without losing the rows
- Find the policy that applied on a given date
- Build an offline voice guide to the handbook
- Evaluate local speech recognition in two languages
- Measure the speed and recall of a vector index
- Build a short meeting brief with source lines
- Train a tiny next-token model from counts
- Train a tiny causal Transformer on the CPU
Tool workflows, retrieval, and local services
Connect a model to tools, review proposed changes, monitor an authenticated server, and test retrieval and coding workflows.
- Expose a read-only tool with MCP
- Connect a local model to the MCP tool
- Review a model’s proposal before saving it
- Require a key for a local model server
- Read queue and generation metrics from a local server
- Find documents across languages
- Rerank retrieved passages with a local model
- Check that a PDF citation still points to the same source
- Train a small classifier and compare it with a chat model
- Ask a local coding model for a change you can test
Documents, media, and serving
Build a PDF handbook workflow, compare local servers and model behavior, and generate speech and images on your computer.
- What attention does inside a language model
- Read dense, mixture-of-experts, and distilled model cards
- Use one local client with several runtimes
- Extract text and page numbers from a PDF
- Build and update a hybrid PDF search index
- Test PDF retrieval and verify cited quotes
- Turn a local text file into speech
- Test whether a vision model reads the right fields
- Measure what happens when requests overlap
- Compare a reasoning model’s thinking settings
- Generate and compare images on your computer
Run, build, and adapt
New lessons cover model types and chat templates, additional setup options, file and audio projects, and testing model changes.
- Choose the right kind of AI for a task
- How chat templates turn messages into tokens
- Run a model with MLX on Apple silicon
- Run on the CPU and check GPU offload
- Use a model on your own second computer
- Stream a local model’s reply
- Build a persistent text search index
- Extract text from a scan with OCR
- Transcribe a recording on your computer
- Store embeddings and search by meaning
- Train and evaluate a LoRA adapter
- Test a local app before updating its model
First edition
The original 16 lessons introduce local inference, model selection, setup, Python, and evaluation.
Planned lessons
Operate a shared inference service
Compare serving runtimes on a Linux GPU, set per-client request limits, and test what happens when the queue fills.
Keep a document index current
Update mixed evidence without rebuilding the full collection. Track deleted sources, changed OCR, and overlapping policies across independent document families.
Make a live local voice interface
Capture a question from a microphone, play the reply incrementally, and handle silence, interruption, and cancellation.
Choose a model with a task-specific evaluation
Compare small and larger models on held-out text, image, and speech examples. Record where quality changes by task, language, or input conditions.