Skip to main content
Back to Library
Prompt Engineering Guide

Mastering Meeting notes extraction
on Llama 3.1 8B

Stop guessing. See how professional prompt engineering transforms Llama 3.1 8B's output for specific technical tasks.

The "Vibe" Prompt

"Extract the key points, decisions, and action items from the following meeting transcript. Make it concise and easy to read. Transcript: [MEETING_TRANSCRIPT]"
Low specificity, inconsistent output

Optimized Version

STABLE
You are an expert meeting assistant for 'Llama 3.1 8B' tasked with extracting structured information from meeting transcripts. Follow these steps meticulously: 1. **Understand the Goal**: The primary goal is to identify and categorize key information into decisions, action items, and general discussion points. 2. **Initial Scan & Context**: Read through the entire transcript once to grasp the overall topics, participants, and flow of the conversation. Note any recurring themes or critical moments. 3. **Identify Decisions**: Systematically scan for explicit statements indicating a choice, agreement, or resolution. Look for phrases like 'We decided to...', 'The team agreed on...', 'It's settled then...', 'Our conclusion is...'. Extract the exact decision. 4. **Identify Action Items**: Systematically scan for tasks assigned to specific individuals or the team, along with deadlines if mentioned. Look for phrases like '[Person] will...', 'We need to...', 'Action: ...', 'Follow up on...', 'Create a draft by...'. Extract the action, the responsible party, and the deadline. 5. **Identify Key Discussion Points**: Extract significant topics, arguments, or information shared that shaped the decisions or led to action items, but are not decisions or actions themselves. Summarize these points concisely. 6. **De-duplicate & Refine**: Review the extracted information for redundancy. Combine similar points. Ensure clarity, conciseness, and accuracy. Remove conversational filler. 7. **Format Output**: Present the extracted information in a structured JSON format with three top-level keys: 'Decisions' (an array of strings), 'Action Items' (an array of objects with 'action', 'assignee', 'deadline' keys), and 'Key Discussion Points' (an array of strings). If a deadline is not specified for an action item, use `null`. If no assignee, use `"Team"`. Meeting Transcript: --- [MEETING_TRANSCRIPT] --- Output in JSON format only:
Structured, task-focused, reduced hallucinations

Engineering Rationale

The optimized prompt leverages Chain-of-Thought (CoT) prompting by breaking down the complex task into smaller, manageable steps. It guides the model through a logical process: understanding, scanning, categorizing, refining, and formatting. This structured approach reduces ambiguity and provides explicit instructions on what to look for and how to present the output. Specifying 'Llama 3.1 8B' in the persona also helps in aligning the model's behavior. The JSON output schema is clearly defined, minimizing hallucination of structure. The detailed instructions for identifying specific entities (e.g., phrases for decisions/actions) and handling missing information (e.g., null for deadlines, 'Team' for assignee) significantly improve extraction quality and consistency compared to the vague 'make it concise and easy to read'.

-250%
Token Efficiency Gain
The output MUST be valid JSON.
The 'Decisions' array MUST contain only strings.
The 'Action Items' array MUST contain objects, each with 'action' (string), 'assignee' (string, or 'Team' if not found), and 'deadline' (string or null).

Ready to stop burning tokens?

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

Optimize My Prompts