CypherMonkeyAI omni-channel lab

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

Learn

Finish in 10 minutes

Read only these opening rooms, export one diagram, and stop.

  1. How this learning lab works — This site is a free learning lab about AI omni-channel agents.
  2. Tokens, not words — A model reads tokens, which may be parts of words.
  3. A job, a shape, and a test — A wish is not a job. Name the reader and the done check.
  4. Why retrieve — A model does not contain your private notes.
  5. An agent is a loop — An agent here means the model can choose a tool, see the result, and continue.
  6. The problem MCP names — Tools written for one app are hard to reuse in another.
  7. Omni-channel, for a learner — Omni-channel here means one assistant pattern that can meet a person on more than one channel.
  8. One GPU, then a cluster — A GPU is a processor often used for the matrix math in model training and serving.
  9. Speech in, a model, speech out — LiveKit's pipeline page names a cascaded pipeline: speech-to-text, then a language model, then text-to-speech.
  10. Transcription of a recording — Wikipedia defines speech analytics as analyzing recorded calls to gather customer information.
  11. Handoff to a person, not a software handoff — A handoff in this lab means a person continues the turn.
  12. Words for this ecosystem — Wikipedia defines speech analytics as analysis of recorded calls.

Short path HTMLShort path PDFShort path PNG

Sources and go deeper

Programming foundations

Files, HTTP, JSON, and a notebook you can re-run. No model call is required to start.

  1. How this learning lab works · L1 · 10 min
  2. Programs, files, and a notebook · L1 · 15 min
  3. Python values and functions · L1 · 20 min
  4. HTTP and JSON · L1 · 20 min
  5. Secrets stay out of the file · L1 · 15 min
  6. Tests and a small eval · L2 · 20 min

Sources and go deeper

Room files: T1-R1 sources.json · T1-R2 sources.json · T1-R3 sources.json · T1-R4 sources.json · T1-R5 sources.json · T1-R6 sources.json

How models work

Tokens, context, sampling, and why a fluent sentence can still be false.

  1. Tokens, not words · L1 · 15 min
  2. Next-token prediction · L1 · 15 min
  3. Context windows · L2 · 15 min
  4. Sampling · L2 · 15 min
  5. Hallucination · L2 · 15 min
  6. Model cards and ids · L2 · 15 min

Sources and go deeper

Room files: T2-R1 sources.json · T2-R2 sources.json · T2-R3 sources.json · T2-R4 sources.json · T2-R5 sources.json · T2-R6 sources.json

Prompting and evals

A job, a shape, and a test. Change one thing and keep the score.

  1. A job, a shape, and a test · L1 · 15 min
  2. Messages · L2 · 15 min
  3. Schemas · L2 · 20 min
  4. A golden set · L2 · 20 min
  5. Latency and a toy budget · L2 · 15 min
  6. Prompt diff · L2 · 15 min

Sources and go deeper

Room files: T3-R1 sources.json · T3-R2 sources.json · T3-R3 sources.json · T3-R4 sources.json · T3-R5 sources.json · T3-R6 sources.json

Retrieval

Find a passage, then answer from it. Measure whether the passage was found.

  1. Why retrieve · L1 · 15 min
  2. Chunks · L2 · 15 min
  3. Embeddings · L2 · 20 min
  4. Retrieve, then read · L2 · 15 min
  5. Cite or refuse · L2 · 15 min
  6. Failure modes · L2 · 20 min

Sources and go deeper

Room files: T4-R1 sources.json · T4-R2 sources.json · T4-R3 sources.json · T4-R4 sources.json · T4-R5 sources.json · T4-R6 sources.json

Agents

A loop that may call a tool, with a schema, a step cap, and a trace.

  1. An agent is a loop · L1 · 15 min
  2. Tools and schemas · L2 · 20 min
  3. Step caps and traces · L2 · 15 min
  4. Human approval · L2 · 15 min
  5. ReAct, at learning depth · L2 · 20 min
  6. When not to use an agent · L2 · 15 min

Sources and go deeper

Room files: T5-R1 sources.json · T5-R2 sources.json · T5-R3 sources.json · T5-R4 sources.json · T5-R5 sources.json · T5-R6 sources.json

Model Context Protocol

A way for a host to talk to tool servers. Confirm the current specification before you build.

  1. The problem MCP names · L1 · 15 min
  2. Host, client, and server · L2 · 15 min
  3. Tools, resources, and prompts · L2 · 20 min
  4. Transports · L2 · 15 min
  5. Consent and roots · L2 · 20 min
  6. A notes server on paper · L3 · 25 min

Sources and go deeper

Room files: T6-R1 sources.json · T6-R2 sources.json · T6-R3 sources.json · T6-R4 sources.json · T6-R5 sources.json · T6-R6 sources.json

Voice and omni-channel agents

One learning model of voice, chat, and messaging, with a handoff to a person. Not a deployment claim.

  1. Omni-channel, for a learner · L1 · 15 min
  2. The voice loop · L2 · 20 min
  3. A latency budget you label as a lesson · L2 · 15 min
  4. Chat and messaging beside voice · L2 · 15 min
  5. Handoff to a person · L2 · 20 min
  6. What this lab will not claim · L1 · 10 min

Sources and go deeper

Room files: T7-R1 sources.json · T7-R2 sources.json · T7-R3 sources.json · T7-R4 sources.json · T7-R5 sources.json · T7-R6 sources.json

Compute, cloud, and a capstone

What a GPU cluster is for, how a cloud API differs, and a small grounded assistant you can actually finish.

  1. One GPU, then a cluster · L2 · 15 min
  2. Colossus as a public story · L2 · 15 min
  3. Cloud endpoints · L2 · 20 min
  4. Logs, regions, and data you may use · L2 · 15 min
  5. A one-page policy · L2 · 20 min
  6. Capstone and certifications · L3 · 25 min

Sources and go deeper

Room files: T8-R1 sources.json · T8-R2 sources.json · T8-R3 sources.json · T8-R4 sources.json · T8-R5 sources.json · T8-R6 sources.json

Voice AI agent stack

Speech in, a model, speech out, and the network that carries the audio.

  1. Speech in, a model, speech out · L1 · 20 min
  2. Turn-taking and barge-in · L1 · 20 min
  3. Latency budgets you label · L2 · 15 min
  4. Telephony and WebRTC · L2 · 20 min
  5. LiveKit Agents and Pipecat · L2 · 20 min
  6. A hosted voice agent, from the xAI docs · L3 · 25 min

Sources and go deeper

Room files: T9-R1 sources.json · T9-R2 sources.json · T9-R3 sources.json · T9-R4 sources.json · T9-R5 sources.json · T9-R6 sources.json

Speech analytics and quality

Transcription, sentiment, a QA score, and redaction, with the pitfalls named.

  1. Transcription of a recording · L1 · 15 min
  2. Sentiment is a label · L2 · 15 min
  3. A QA score needs a rubric · L2 · 20 min
  4. Redaction · L2 · 20 min
  5. Pitfalls · L3 · 20 min

Sources and go deeper

Room files: T10-R1 sources.json · T10-R2 sources.json · T10-R3 sources.json · T10-R4 sources.json · T10-R5 sources.json

AI agent safety

Handoff, guardrails, consent as education, and a test you can re-run.

  1. Handoff to a person, not a software handoff · L1 · 15 min
  2. Guardrails · L2 · 20 min
  3. Consent and recording, as education · L2 · 20 min
  4. Evaluation and testing · L2 · 20 min
  5. What this chapter will not claim · L3 · 15 min

Sources and go deeper

Room files: T11-R1 sources.json · T11-R2 sources.json · T11-R3 sources.json · T11-R4 sources.json · T11-R5 sources.json

Contact-center ecosystem

A glossary, a mind map, sister systems, an integration map, bridges, and future hooks.

  1. Words for this ecosystem · L1 · 15 min
  2. A mind map · L1 · 15 min
  3. Sister technologies · L2 · 20 min
  4. An integration map · L2 · 20 min
  5. Bridges across platforms · L2 · 20 min
  6. Future hooks · L3 · 15 min

Sources and go deeper

Room files: T12-R1 sources.json · T12-R2 sources.json · T12-R3 sources.json · T12-R4 sources.json · T12-R5 sources.json · T12-R6 sources.json