🧠 Agentic Research Abstract Generator & Web Summariser-2

An agent-based system using LangGraph and HuggingFace models to generate research abstracts and summarize web content, with modular workflows, memory handling, and task-specific agents.

This builds upon the previous repo

• Project Link
• ReadyTensor-Publication

🚀 Features Improvements

• 📈 Evaluation metrics and logging
• 📈 Evaluation metrics and logging
• 📈 Testing features incorporated
• 🚀 Model deployment with streamlit

How the project goes

• All files can only work after installing all dependencies in the environment.yml file
• The notebooks folder contains the jupyter notebook file for testing the project as a whole and for experimenting. research_graph2.ipynb contains the experimentation for abstract generation while research_graph3.ipynb contains the experimentation for web content summarisation.
• The graph_article folder contains the critic, writer and graph_article python files. The writer.py(writer agent) file takes the category and title needed for drafting the abstract while the critic.py (reviewer agent) reviews the generated abstract. The graph_article.py connects boths writer.pyand critic.py by using LangGraph.
• The graph_web folder contains the grap_web, loader, search and summarizer python files. The search.py (search agent) file searches takes in the URL link for web search, while the loader.py (loader agent) loads the web page but limits it to appoximately 32,000 tokens to not exceed the max token limit. The summarizer agent in summarizer.py files, provides a concise summary for the the URL given. The graph_web.py connects all components together as one.
• The utils folder contains the visualizer.py file which creates the graphs of the grah_web.py and graph_article.py files when called in main.py. The generatd graphs are saved to the visuals folder.
• The shared.py file contains the shared state for the summarizer and abstract generator graphs.
• The main.py file calls all graphs together and prompts the user for if the would like to generate an abstract or summarise a webpage.
• The img folder contains the images used for visualisation, also Langsmith_run.png show the an example run when Langsmith is used for tracing the graph.
• The tests folder contains all necessary tests for the each component for proper integration of the project, making it ready for deployment.

Replicating this project and Example Usage

• Kindly refer to the GitHub link for the previous project for replication.
• Run pip install pytest>=8.4 and pip install pytest-mock==3.14 after using the environment.yml from the previous repo

Example Streamlit Outputs in Development

Example Streamlit Outputs in Production

NB:

• You must have a huggingface api key to use the streamlit app. Kindly refer to the previous github repo to know how to generate a huggingface api key

Here is the Publication on;

• ReadyTensor

Issues faced:

• Integrating tests for the LLM agents.
• Conflicts when deploying app on Streamlit, hence the environment.yml file was removed.
• Selenium was hard to incorporate for deployment, hence beautisoup was used for an attempt to read the web data.