Agents Getting Real Things Done - Breakthrough - Combine RPA with CREWAI or Autogen and GPTVision

---

---

AI Summary

Video Summary

Workflow Overview

• The video presents a workflow combining an agentic framework (Qi or Autogen) with Microsoft Power Automate for personalized outreach.
• The creator is implementing this workflow in their real business due to the control and validation layers provided by the RPA tool.
• The goal is to automate outreach campaigns with high-quality, personalized messages using data from platforms like Facebook, LinkedIn, or Instagram.

Tools Used

• Agentic framework (Qi for simplicity, Autogen for more features)
• Microsoft Power Automate Desktop
• Excel as a database
• OpenAI’s GPT-3.5 and GPT-4 for text generation and image analysis

Workflow Steps

1. Scrape leads from social media platforms using data scrapers and export to CSV.
2. Copy profile URLs to a master Excel file.
3. Use GPT-4 Vision to analyze cover photos and scrape profile text and recent posts.
4. Aggregate data and use GPT-4 to craft personalized outreach messages.
5. Send the message to Qi agents for scoring based on parameters like professionalism, offensive language, and reputational risk.
6. Use the scores to filter out messages and proceed with high-scoring ones.

Automation Process

• The automation involves launching Excel, nullifying variables, and looping through prospects’ URLs.
• A JavaScript clicks “See More” buttons on Facebook posts to reveal full content.
• GPT-4 Vision is used to describe cover photos.
• The prospect’s name, intro, and recent posts are fetched and aggregated.
• GPT-4 generates personalized outreach messages based on the aggregated data.
• Qi agents score the messages, and high-scoring ones are considered for sending.

Challenges and Solutions

• The creator faced challenges with sending API requests due to formatting issues with JSON.
• A tool was used to validate JSON and a Python script to sanitize strings for JSON compatibility.

Future Improvements

• Refine prompts for GPT-3.5 to improve message personalization.
• Make Qi agents more critical in their scoring.
• Add more parameters for scoring.
• Automate the direct messaging process.

Conclusion

• The creator is excited about the potential of this workflow to remove bottlenecks in their automation process.
• The workflow allows for personalized outreach at scale, with control over the content to protect one’s reputation.

Detailed Instructions and Tips

• No specific CLI commands, website URLs, or detailed instructions were provided in the summary.