CypherMonkeyAI omni-channel lab

A lifelong-learner lab on AI omni-channel agents, including voice and the contact center. Free to read. Nothing sold.

Lifelong-learner lab

Learn AI omni-channel agents without a job claim.

Programming, models, retrieval, agents, MCP, voice and messaging, a voice stack, speech analytics, agent safety, a contact-center map, and a cloud or cluster story you re-check. This is a free learning lab. It does not claim professional GenAI work.

Finish in 10 minutesThe short path only. Full course70 rooms across 12 tracks. Just downloadDiagrams, sheets, cards, and a zip.
One policy, three channelsPersonPolicyVoiceChat

Tracks

Programming foundationsFiles, HTTP, JSON, and a notebook you can re-run. No model call is required to start.How models workTokens, context, sampling, and why a fluent sentence can still be false.Prompting and evalsA job, a shape, and a test. Change one thing and keep the score.RetrievalFind a passage, then answer from it. Measure whether the passage was found.AgentsA loop that may call a tool, with a schema, a step cap, and a trace.Model Context ProtocolA way for a host to talk to tool servers. Confirm the current specification before you build.Voice and omni-channel agentsOne learning model of voice, chat, and messaging, with a handoff to a person. Not a deployment claim.Compute, cloud, and a capstoneWhat a GPU cluster is for, how a cloud API differs, and a small grounded assistant you can actually finish.Voice AI agent stackSpeech in, a model, speech out, and the network that carries the audio.Speech analytics and qualityTranscription, sentiment, a QA score, and redaction, with the pitfalls named.AI agent safetyHandoff, guardrails, consent as education, and a test you can re-run.Contact-center ecosystemA glossary, a mind map, sister systems, an integration map, bridges, and future hooks.

What this lab means by an agent

An omni-channel agent, here, is one policy that can meet a person on voice, on chat, or in a message, and can hand that person to a human. This site is a free learning lab. It does not claim a deployment, a job, or a certificate.

The NIST AI Risk Management Framework is the vocabulary this lab uses for that risk. It is not a product certificate. The skip link, the menu, and the type-size control follow the intent of W3C, Web Content Accessibility Guidelines 2.2. Web accessibility — Wikipedia is the short public overview of that idea.

A model continues text

Vaswani and co-authors, Attention Is All You Need, 2017 describes the transformer architecture later language models build on. Brown and co-authors, Language Models are Few-Shot Learners, 2020 reports that a large model can follow a few examples written in the prompt. A fluent sentence can still be false. The check in this lab is the source, not the tone. Large language model — Wikipedia and Attention Is All You Need — Wikipedia are the public summaries.

Retrieve, then answer

Lewis and co-authors, Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, 2020 puts retrieved passages into the generation step. The rule on this site is shorter: if the note is silent, refuse. Fluency is not a citation. Retrieval-augmented generation — Wikipedia is the public summary of that pattern.

A loop is not a workflow

Yao and co-authors, ReAct: Synergizing Reasoning and Acting in Language Models, 2022 interleaves a reason step and a tool step. Anthropic, Building effective agents is the vendor essay this lab uses when it separates a fixed workflow from a loop that chooses a tool. Keep a schema, a step cap, and a trace. A person approves an action that is hard to undo.

One way to name tools

Model Context Protocol specification, 2025-06-18 names a host, a client, and a server. What is the Model Context Protocol (MCP)? is the public introduction. Model Context Protocol — Wikipedia is the encyclopedia summary. Confirm the current revision before you build. This page opens no port.

The older campus shell, with 593 study pages, stays available as prior notes at campus/ai-lab-university.html.

Sources and go deeper