This repository showcases various examples of Language Model (LLM) agents, demonstrating their capabilities in different domains.
LLM Agents are AI systems that use large language models to understand and generate human-like text, combined with the ability to reason about problems and take actions. These agents can break down complex tasks, make decisions, and execute actions based on their understanding of the input and the context.
One of the key frameworks used in this repository is ReAct (Reason + Act). ReAct is an approach that enhances the problem-solving capabilities of language models by interleaving reasoning and acting. Here's a brief overview:
- Reason: The agent analyzes the problem, considers possible approaches, and plans a course of action.
- Act: Based on its reasoning, the agent takes an action, such as using a tool or generating a response.
- Observe: The agent observes the results of its action.
- Repeat: The process repeats, with the agent reasoning about the new state and deciding on the next action.
This cycle allows the agent to tackle complex, multi-step problems by breaking them down into manageable pieces and adapting its approach based on intermediate results.
- Multi Agents Clinical Metadata Extraction: A multi agentic system for extracting and structuring clinical metadata from medical case studies. More info on dataset can be found here.
- The
RetrieverAgent
is used to extract relevant text chunks for a subset of the metadata categories. - The
MetadataExtractionAgent
is then used to generate the final structured metadata, leveraging the text chunks retrieved by theRetrieverAgent
. - The
OutputValidator
is used to ensure the generated metadata conforms to the expected data model.
- The
- Natural Language to SQL Agent: An NL2SQL agent that leverages Large Language Models (LLMs) and the ReAct (Reason+Act) framework to convert natural language queries into SQL and perform various database operations.
Each example demonstrates how LLM agents can be applied to specific domains, showcasing their flexibility and power in solving real-world problems.
pip install -r requirement
Then follow the instruction given within the notebooks to run the agent.