What predictive AI, generative AI and LLMs actually do — then how Red Hat turns models, agency knowledge and automation into an operable mission capability, proven in eight labs you run yourself.
An hour where we run the seven labs together, show and tell, and find the point in the stack where you want to prove value. That scopes the pilot.
A half-day with your platform and security teams to size accelerators, place the first workload and name the data classes in scope.
One workflow, on your infrastructure, with a measured baseline and an agreed definition of success before we start.
Four Inc. and Carahsoft carry the vehicles and pricing — bring them in early, not at the end.
Red Hat · if nothing else, he will get you to the right person
bscalio@redhat.comIRS account executive: Ted Craig, Red Hat. Everyone else, email Brad and he connects you to yours.
Red Hat
FOURinc.
carahsoft
Everything from today — deck, labs, guides — at nommsweymx.github.io/redhat-ai-tax-labs. The seven labs are the commands your engineers will run on day one. Follow it on your own: …/adventure.html.
the next decision is yoursEight labs, roughly five minutes each (Lab 08 runs the whole loop). Press Show to reveal the next command and type it yourself, or press Run to execute it for you. Commands are real and copyable; output is representative of a working environment so the lab runs the same for everyone. Demo plays a whole lab hands-free — and the Demo view in the sidebar runs all seven on a loop. Not sure where to start? The choose-your-own-adventure page lays out a reading order for every job in the room. As each command runs, a card pops up under the terminal: what is happening, why it matters to this talk, and the seam a single platform removes. Every lab also carries a theory panel explaining what actually happened, and the glossary at the bottom demystifies the jargon.
The small piece of text a model actually reads and writes — roughly four characters. “Form 8822” is three tokens; a page of a notice is about 700.
Splitting text into tokens and turning each one into a number — the only translation step between your words and the mathematics.
The billions of numbers learned during training. They encode patterns of language, not a database of facts, and they are frozen once deployed.
Running the model: tokens in, one predicted token out, repeated until the answer is complete. Lab 02 is the only lab that changes weights — every model call in the others is inference.
Everything you send — the instructions plus the text to work on. Sending it is the whole request; no other step exists.
How much the model can consider at once: your prompt plus its own answer so far. Short-term memory, cleared after every request.
How adventurous the next-token choice is. Near zero the model repeats its best guess reliably — the right setting for tax work.
Text converted into a vector — a point in space where similar meanings sit close together. What makes searching by meaning possible.
A database of embeddings. Labs 04 and 07 run one in pgvector so guidance and logs can be found by meaning rather than by keyword.
Fetch the relevant passages from your documents, place them into the prompt, and require the answer to cite them.
Deliberately adjusting weights with your own examples. LoRA trains a small add-on instead of all the weights — cheaper, faster, reversible.
A confident answer assembled from plausible tokens with nothing behind it. Countered with retrieval, thresholds, refusal — and a human signature.
The input mix changing compared to the baseline you accredited. It means the world changed — not that the model decayed. Weights cannot rot.
Bring the reference architecture from slide 9 to your own platform and security teams in a half-day workshop.
Contract vehicles, pricing and the paperwork path for a pilot that has to start this fiscal year.
Procurement support across the agencies and integrators already running Red Hat platforms.
One thing to remember. Nothing in these seven labs required a public model endpoint, an internet connection at inference time, or a rewrite of a system of record. That is the whole argument.
Space plays and pauses · ← → switch labs · loops through all eight continuously. A card beside the terminal explains each step as it runs. Open the page with #demo on the URL to start here, playing.