How It Works

Understand the journey from spoken discussion to structured knowledge atoms.

This page explains what happens between your seminar discussion and the knowledge that appears in your workspace.

The Journey: Voice to Knowledge

Your classroom discussion goes through three major transformations:

  1. Spoken discussion — what you say in the seminar
  2. Transcribo transcript — an AI-generated written record of who said what, and when
  3. Knowledge atoms — individual ideas extracted, classified, and connected

The 8-Stage Pipeline

Behind the scenes, the transcript passes through eight stages:

Stage 1: Read the Transcript

The system reads the written record of your session — a Markdown file produced by Transcribo that captures each speaker's words with timestamps.

Stage 2: Parse Speaker Turns

The transcript is broken into individual speaker turns. Each turn records who spoke, what they said, and when.

Stage 3: Anonymize Speakers

Real names are replaced with anonymous labels to protect privacy. The instructor is labeled Instructor, and students become Speaker 1, Speaker 2, Speaker 3, and so on. The mapping between real names and labels is stored securely and never appears in the knowledge base.

Stage 4: Extract Ideas

This is the core of the pipeline. An AI model reads through the anonymized transcript and identifies discrete ideas — contributions that move the group's thinking forward. A single idea might span multiple speaker turns, and one speaker turn might contain several ideas.

Small talk, greetings, and pure agreement ("yeah, good point") are filtered out. What remains are the substantive contributions.

Stage 5: Detect Relationships

The system identifies how ideas relate to each other. Does one idea build on another? Does it challenge a previous claim? These connections are captured automatically.

Stage 6: Write a Session Log

A human-readable Markdown log is generated for each session, listing every extracted idea with its classification and speakers. This log serves as a record the instructor can review.

Stage 7: Store in PrivateLanguage

Each idea becomes a knowledge atom in your team's workspace. Atoms include the idea text, its classification, contributing speakers, timestamps, and topic tags.

Stage 8: Record Relationships

The detected relationships between ideas are stored, so you can trace how one insight connects to another.

Speaker Anonymization

Privacy is a core design principle. When the pipeline processes a transcript:

In the transcriptIn the knowledge base
Chris MooreInstructor
First student to speakSpeaker 1
Second student to speakSpeaker 2
Third student to speakSpeaker 3

Speaker numbers are assigned in the order people first appear in the transcript. The same person keeps the same number throughout a session.

The Six Action Types

Every extracted idea is classified by what it does to the group's thinking:

Action TypeWhat It MeansExample
ProposalIntroduces a new idea or direction. Expands the possibility space."What if we approached this through the lens of dual coding theory?"
Build-onExtends or refines someone else's idea. Advances a path already started."Building on that, the act of choosing what to draw forces you to prioritize."
CritiqueChallenges an idea or identifies problems. Constrains the possibility space."But doesn't that mean you might miss important details?"
ConstraintSurfaces a hard limit the group must respect. A boundary condition."We only have three weeks, so we can't do a full longitudinal study."
QuestionOpens an inquiry that redirects the group's attention."How do concept maps relate to what we're doing here?"
DiscoveryA surprise finding or hidden connection revealed. Something nobody expected."Wait — that's the same pattern we saw in the ethnography readings."

These types capture the epistemic function of each idea — what it does to the group's collective knowledge state, not just what it says.

Scope Tags

Ideas are also tagged with a scope — a short topic label that groups related ideas together. For example:

  • visual-thinking — ideas about visual representation and cognition
  • dual-coding — ideas related to Paivio's dual coding theory
  • cross-cultural — ideas about cultural differences in knowledge practices
  • timeline — ideas about temporal structures and sequencing

Scope tags use kebab-case (lowercase words joined by hyphens). Ideas sharing the same scope tag are likely related, making it easy to find clusters of thinking around a particular topic.