Setting up three GPT research agents using OpenAI API Playground. This involved making them communicate together to carry out a deep web scrape research analysis and yield results which are more comprehensive than surface level web search. The research questions and results were read and written on an AirTable database using their API.
Three GPT agents were deployed in a heighrachical relationship:
- The Director
- The research Manager
- The researcher
This hierarchy enabled much deeper results than normal. Especially considering the research manager’s role, who denied any of the researcher’s results which weren’t satisfactory in the meeting the target research objectives.
This was implemented using:
- Python 3 (development environment)
- OpenAI Playground API
- GPT4-Preview (GPT model)
- AssistantAPI (inc. fine-tuning GPT agents and custom instructions)
- AutoGen (framework to connect LLM agents together enabling multi-agent conversations)
- Langchain (framework to connect LLM agents together)
- Beautiful soup (Python package for web scraping)
- Serper.dev (Google search functionality)
- Browserless.io (API web scraping service)
Overall it was successful and was able to provide a much deeper web scrape and research analysis than simply one GPT agent.