2 min read

AI Multi-Agents for Everyone

AI Multi-Agents for Everyone

In recent years, the rise of ChatGPT has captured global attention, with many leveraging it for various tasks and some exploring alternatives from tech giants like Google and Facebook. However, what is often overlooked is that ChatGPT or any other large language model (LLM) is not a standalone AI solution. These models are akin to microchips in a computer—essential components, yet requiring integration with various other LLM-based agents and external APIs to function as a truly useful AI assistant.

At Integrail, we have spent over three years experimenting with LLMs, refining numerous approaches. Today, we are excited to share our insights and best practices to help you build effective AI agents for personal or business use.

First, What is LLM?

An LLM, including those labeled as “multi-modal,” is fundamentally an artificial neural network (ANN) with a transformer architecture. It processes a sequence of characters and predicts the next symbol (or token) probabilistically based on its training data. While this mechanism is surprisingly effective, it has significant limitations.

Pretty much every LLM out there

To make LLMs truly useful, additional elements are essential: 

Improved LLM
  1. Vector Memory: This involves searching and retrieving relevant context based on the semantic meaning of the text, not just keywords. This technique, recently termed “retrieval-augmented generation” (RAG), enhances contextual understanding.
  2. Fine-Tuning: Additional training on top of the base model helps the LLM better understand specific data sets, tones of voice, and other nuances.
  3. Superior Training Data: If budget allows, using better training data to retrain smaller-parameter models from scratch can sometimes yield performance that surpasses even advanced models like GPT-4.

Not Just ChatGPT

Another key point is that you are not forced to use OpenAI models for all of your tasks. In fact, on some specialized tasks (such as ubiquitous entity extraction, which you need for pretty much any multi-agent system) much smaller fine-tuned Llama-7B model outperforms GPT-4!

In our tools, we provide easy ways to compare performance across the most popular models, e.g. the screenshot above shows a typical entity extraction in JSON task execution by 5 different models in parallel.

So what is a Multi-Agent then?

To understand the concept of a multi-agent system, let's consider the task of creating a comprehensive analytics report on the USA economy. This task requires far more than a single query to ChatGPT. A functional AI assistant needs to:

  • Gather diverse sources on the USA economy through web searches
  • Interact with users for additional information
  • Analyze and understand the collected data
  • Compose the report section by section
  • Include relevant charts and tables
  • Format the final document appropriately

All of it is possible, but using multi-agent architecture, where each subtask is being executed by a specific agent, based on an LLM or other ANN with additional ability to call external APIs (e.g., for web search or converting html text to something more readable etc.).

That is exactly what we do at integrail.ai — next time we’ll dive deep into specific AI multi-agents design and how easy it is to do using our platform, but for now you are welcome to register and start playing! Oh, and while we are in beta, everything is FREE! :)

What is LLM Benchmarking?

What is LLM Benchmarking?


Read More
Retrieval Augmented Generation (RAG) In 2 Minutes

Retrieval Augmented Generation (RAG) In 2 Minutes

Creating "pragmatic AGI" (artificial general intelligence) with current technology is achievable without waiting for future advancements like GPT-5...

Read More
How Multi-Agent Systems Benefit Startups

How Multi-Agent Systems Benefit Startups

Introduction Startups constantly seek innovative solutions to gain a competitive edge. Multi-agent systems in AI, such as those provided by...

Read More