Generate structured meeting minutes from raw transcripts
A multi-step AI prompt system designed to turn meeting transcripts or notes into clear, consistent meeting minutes.
Use case
This system is designed for situations where a meeting transcript or set of notes needs to be transformed into a clear, formal, and usable meeting record.
It works particularly well when:
transcripts are informal, imprecise, or generated automatically
notes are fragmented or taken quickly during the meeting
accuracy matters, but you still want to avoid dedicated and expensive external tools
If the meeting is short or does not require full consolidation, the system can be used with only the first agent.
For longer or more complex meetings, a second agent can be used to consolidate and finalize the document.
System components
This system is built around two distinct agents, each with a specific role.
Agent 1 - Enrichment & Rewriting
Takes chunks of the meeting transcript or notes (recommended: up to ~2,000 words at a time)
Treats the input as oral, imperfect transcription
Actively compares the text with reference documents provided by the user
Corrects language, clarifies concepts, and aligns terminology and data
Rewrites the content into a clear, formal, and self-contained text section
Inputs
Meeting transcript or notes (pasted manually, in chunks)
Optional reference documents related to the meeting topic
Knowledge base
Required - provided as a structured template that the user fills in before running the system
Agent 2 - Final consolidation (optional)
Takes the full document obtained by merging the outputs of Agent 1
Reviews the text globally
Deduplicates repeated concepts
Harmonizes terminology and style
Produces a final, coherent meeting document
Input
Outputs of Agent 1
Knowledge base
The same used in Agent 1
This agent is optional and is used only when a global revision or synthesis is needed.
How the system works
Process
Obtain a meeting transcript or written notes
(for example from tools like Teams, Google Meet, Zoom, or manual notes)Split the text manually into chunks of up to ~2,000 words
Run Agent 1 on each chunk, using the provided reference documents and knowledge base
Manually merge the rewritten sections into a single document
(Optional) Run Agent 2 to consolidate and finalize the full meeting minutes
Each step is intentionally explicit to maintain control and accuracy.
Why two agents?
The system uses two separate agents because the two steps require different types of reasoning.
The first agent works locally, chunk by chunk.
Its job is to stay faithful to the original content, correct inaccuracies, and enrich the text using reference documents - without losing detail.The second agent works globally.
Its job is to look at the document as a whole, remove repetitions, harmonize terminology, and produce a coherent final version.
Trying to do both things with a single agent increases the risk of:
premature summarization
loss of detail in early steps
inconsistent structure or terminology
Keeping the agents separate makes the system more reliable, easier to control, and easier to evolve over time.
Who this system is for
This system is useful if you:
work with meeting transcripts or notes
need formal, structured meeting minutes
care about accuracy and consistency
prefer manual control over automated black-box tools
Links to install system components
Available with access
Instructions & setup notes
Step-by-step guidance on how to set up the system, run each agent, handle documents, and avoid common pitfalls.Agent 1 - Enrichment & Rewriting
Prompt and rules for processing transcript chunks, enriching content with reference documents, and producing clean, formal text.Agent 2 - Final consolidation (optional)
Prompt and rules for reviewing the full document, deduplicating content, and producing the final meeting minutes.Knowledge base template
A ready-to-fill template defining structure, style rules, and reference material used by the system.