L'agent

An AI-powered solution to revolutionise the way organisations manage information. AI agents + a hybrid RAG ensure that information reaches the right people, with actionable insights.

Comment

Project history: L'agent, an information management system powered by AI agents

Inspiration 💡

We have experienced in our professional lives that the management of incoming information is slow and time consuming, getting the right information to the right person and using it intelligently for next actions. We wanted to solve this problem with the help of AI agents & a hybrid RAG system!

What it does 🎈

L'agent is an AI-powered solution designed to revolutionise the way organisations manage information. It intelligently analyses, prioritises and distributes incoming information using a hybrid RAG system and a series of specialised AI agents. This ensures that the right information reaches the right people, with actionable insights provided for high priority items.

How we built it 🔧

  1. Hybrid RAG System Integration: We started with the development from scratch of our Hybrid RAG system, which combines a traditional RAG system, a 'smart' search in the vector DB and a 'home-made' re-ranking using Mistral Large. This system is used by the Dispatch Agent to decide who should be informed about the information received.

  2. AI agent development: We then integrated individual AI agents: the routing agent, the analyst agent, the dispatch agent and the strategic agent. Each agent was trained on a specific dataset and task.

  3. Information flow pipeline: Once the agents were ready, we created a pipeline for the information to flow through. This involved defining the rules for how the agents would handle incoming information in order to create a coherent workflow between our agents.

  4. User Interface: Finally, we developed a user-friendly interface that allows users to monitor the system and see how information is managed.

Challenges we ran into 🌉

  1. Scope & coordinate the AI Agents: One of the biggest challenges we faced was defining the right scope for the different AI agents and ensuring that they were able to work together. It was a complex task to ensure that information was passed correctly between agents and that the system as a whole was efficient and effective.

  2. Developing the hybrid RAG system: Developing our hybrid RAG system was also a challenge. We had to find the right balance between the more traditional RAG system and our AI-based model to ensure the system was accurate and efficient. We preferred not to use an existing library that could facilitate this, as we prefer to learn and master each step of the optimisation process if necessary.

  3. Actionable insights generation: Developing the strategic AI agent that generates action plans was a complex task. We had to ensure that the action plans were relevant, useful and actionable.

  4. User Interface: Building the right user interface for this use case with limited time.

Accomplishments that we're proud of 🏆

  1. Successful coordination of AI agents: We're proud of our successful coordination of multiple AI agents, resulting in a seamless information management system.

  2. Hybrid RAG system: We're also proud of our innovative self-made hybrid RAG system, which significantly improves the accuracy and efficiency of information dispatching.

  3. Actionable Insights: We're excited that our system doesn't just manage information, but also provides actionable insights adapted to the person that receives it, making it a valuable tool for any organisation.

What we learned 🧠

  1. AI Agent scoping & coordination: We have gained a deep understanding of how to scope & coordinate multiple AI agents, each with a unique role, to create a seamless information management system.

  2. Hybrid RAG systems: We challenged our knowledge of the benefits of combining traditional RAG systems with AI-powered models, such as our home-grown "re-rank" powered by Mistral Large, to improve the accuracy and efficiency of information dispatch.

  3. Actionable Insights: We discovered the importance of not only managing information, but also providing actionable insights. This led to the development of our Strategic AI Agent, which generates action plans for high-priority information.

What's next for L'agent 🌐

We plan to release everything open source on our GitHub. We want to refine and optimise our AI agents and the hybrid RAG system. We also intend to incorporate user feedback to improve the user interface and overall functionality of L'agent. If possible, we would like to partner with a company to test it in a live environment. Our goal is to make L'agent the information management solution of choice for organisations worldwide. 🤗

Built With

Share this project:

Updates