Skip to main content
Back to Library
Prompt Engineering Guide

Mastering Meeting notes extraction
on Llama 3.1 405B

Stop guessing. See how professional prompt engineering transforms Llama 3.1 405B's output for specific technical tasks.

The "Vibe" Prompt

"Extract the key decisions, action items, and participants from these meeting notes. Meeting Notes: [MEETING_NOTES_TEXT]"
Low specificity, inconsistent output

Optimized Version

STABLE
You are Llama 3.1 405B, an advanced AI designed for precise information extraction and summarization. Your task is to meticulously extract specific entities from raw meeting notes. Follow a step-by-step thinking process to ensure accuracy and completeness. **Meeting Notes Analysis Plan:** 1. **Identify Core Meeting Content:** First, read through the entire meeting notes to understand the overall context, main topics discussed, and general flow of the conversation. Note any sections that seem to delineate different discussion points. 2. **Decision Extraction:** Systematically scan the notes for explicit statements indicating a decision was made. Look for keywords or phrases such as 'decided to', 'agreed upon', 'the team concluded', 'it was resolved that', 'finalized', 'will proceed with'. Extract the full, clear statement of each decision made. 3. **Action Item Extraction:** Next, identify all action items. These are tasks that need to be completed by a specific person or group by a certain deadline. Look for phrases like 'X will do Y', 'Z is responsible for', 'need to follow up on', 'to be completed by', 'assigned to'. For each action item, clearly identify: * The action to be taken. * The person(s) responsible (if stated). * The due date (if stated). 4. **Participant Identification:** Scan the notes for names of individuals who spoke, were mentioned as attending, or were assigned tasks. Compile a unique list of all identified participants. Do not include generic roles unless they are distinct entities (e.g., 'the marketing team' if no individual names are present, but prefer 'John Doe' over 'the marketing lead' if John Doe is also mentioned). 5. **Output Formatting:** Present the extracted information in a structured JSON format as specified below. Ensure all extracted items are concise and accurate reflections of the original notes. **Meeting Notes:** [MEETING_NOTES_TEXT] **Extracted Information Output Format (JSON):** ```json { "decisions": [ "[Decision 1 statement]", "[Decision 2 statement]" ], "action_items": [ { "action": "[Description of action]", "responsible_person": "[Name of person or team, or 'N/A']", "due_date": "[Date, or 'N/A']" }, { "action": "[Description of action]", "responsible_person": "[Name of person or team, or 'N/A']", "due_date": "[Date, or 'N/A']" } ], "participants": [ "[Participant 1 Name]", "[Participant 2 Name]" ] } ``` **Thought Process:**
Structured, task-focused, reduced hallucinations

Engineering Rationale

The optimized prompt leverages several techniques to enhance performance for Llama 3.1 405B: 1. **Role Assignment & Persona:** 'You are Llama 3.1 405B, an advanced AI...' primes the model for high-quality, precise output consistent with its capabilities. 2. **Chain-of-Thought (CoT):** The 'Meeting Notes Analysis Plan' explicitly outlines a step-by-step thinking process. This guides the model to break down the task, reducing the cognitive load and ensuring a systematic approach to extraction. It prevents the model from jumping directly to an answer. 3. **Specific Instruction for Each Entity:** Each extraction type (decisions, action items, participants) has dedicated instructions, including keywords to look for and specific details to capture (e.g., action, responsible person, due date for action items). 4. **Negative Constraints/Clarifications:** 'Do not include generic roles unless they are distinct entities' helps prevent common errors in participant extraction. 5. **Strict Output Format:** Providing a precise JSON schema with example values minimizes ambiguity about the desired output structure, making it easier for the model to generate parseable JSON. 6. **Explicit 'Thought Process' Placeholder:** The 'Thought Process:' at the end encourages the model to output its reasoning (if enabled to do so), which can be useful for debugging or understanding its extraction logic. 7. **Clarity and Conciseness:** While longer, the prompt is highly structured and clearly articulated, reducing misinterpretations compared to a vague request.

0%
Token Efficiency Gain
The output JSON should correctly identify all explicit decisions made in the meeting notes.
Each identified action item in the JSON should include the action, responsible person (if present), and due date (if present).
The 'participants' array in the JSON should list all unique individuals mentioned as attending or participating.

Ready to stop burning tokens?

Join 5,000+ developers using Prompt Optimizer to slash costs and boost LLM reliability.

Optimize My Prompts