AI-Assisted Knowledge Synthesis

Three CTA sessions with a senior expert produce forty to sixty pages of transcripts, dozens of scenario notes, and multiple rounds of probe responses. The synthesis challenge — organizing this material into a usable knowledge artifact — is where most extraction efforts lose the value they extracted. AI-assisted synthesis is not about automating the extraction. It is about accelerating the sense-making that turns raw transcript into structured knowledge.

1. Transcript Preparation Clean and label each transcript with metadata: expert name, session number, task domain, date, technique used. Structure verbatims by phase — task decomposition, knowledge probe, scenario response. The metadata allows the synthesis model to distinguish between high-confidence explicit knowledge and low-confidence tacit inference.
2. Pattern Extraction Prompting Feed labeled transcripts to an LLM with structured synthesis prompts: "Identify the decision criteria mentioned across all sessions." "List the conditions the expert names as changing their approach." "Identify the cues they describe noticing that they do not believe novices would notice." These structured prompts extract the conceptual layer that transcript reading alone would take days to surface.
3. Knowledge Gap Detection Use the synthesized model to identify what is still missing. Prompt: "What questions remain unanswered about this task domain based on the transcript material?" "What decision points appear underspecified?" "Where does the expert give contradictory accounts across sessions?" The gaps become the agenda for follow-up sessions.
4. Structured Knowledge Draft Output the synthesis as a structured draft: decision trees for judgment calls, condition tables for contextual variation, annotated process flows for procedural knowledge, and a explicit list of the tacit rules the expert demonstrated but did not state. This draft goes back to the expert for validation — not as final output, but as a prompt for correction.

SYNTHESIS PROMPT PATTERN — Expert Knowledge Extraction

Input: [Expert Name] session transcripts, sessions 1-3
Task domain: [Domain]

Extract the following from the transcript material:

1. DECISION CRITERIA
   List every condition the expert names as a factor in their judgment.
   Format: "When [condition], expert does/considers [action/factor]"

2. CUES AND SIGNALS
   List every cue or signal the expert describes noticing.
   Mark each as: EXPLICIT (stated rule) or TACIT (demonstrated but not stated rule)

3. CONTEXTUAL VARIATION
   List every "it depends" statement and enumerate the conditions
   that trigger different expert behavior.

4. COMMON FAILURE MODES
   List every mistake, near-miss, or novice error the expert describes.

5. KNOWLEDGE GAPS
   List every decision point where the expert's account is:
   - Incomplete (step assumed rather than stated)
   - Contradictory across sessions
   - Too general to be actionable
   → These gaps become the agenda for Session 4.