Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1982

Automating Meeting Management: An AI-Driven

Web-Based Meeting Management System
Vaibhavi Ladhe1; Suraj Magdum2; Dhruv Mahajan3; Eeshan Malwandikar4;
Nikhil Dhavase5
1,2,3,4,5
Information Technology, MMCOE, Savitribai Phule Pune University, Karve Nagar, Pune, India

Publication Date: 2025/05/13

Abstract: The Meeting Manager Application is a web-based tool designed to simplify the organization, execution, and
documentation of meetings. The application offers a comprehensive suite of features, including the ability to create meetings,
upload relevant documents, manage discussion points, generate agendas, transcribe meeting recordings, and produce
detailed meeting summaries. It leverages advanced technologies such as Flask for the web framework, SQLAlchemy for
database management, ChromaDB for vector storage, and Google’s Gemini Large Language Model (LLM) for natural
language processing tasks. The system integrates state-of-the-art machine learning techniques, utilizing Sentence
Transformers to generate text embeddings, which are stored and queried for similarity search in ChromaDB. By automating
key aspects of meeting management, the application improves efficiency and reduces manual effort. This project
demonstrates how AI-driven solutions can streamline workflow processes in corporate settings. Future enhancements
include support for multi-user collaboration, scalability, and real-time meeting analysis for enhanced user experience.

How to Cite: Vaibhavi Ladhe; Suraj Magdum; Dhruv Mahajan; Eeshan Malwandikar; Nikhil Dhavase (2025). Automating Meeting
Management: An AI-Driven Web-Based Meeting Management System. International Journal of Innovative Science and Research
Technology, 10(4), 3358-3364. https://doi.org/10.38124/ijisrt/25apr1982

I. INTRODUCTION The Meeting Management System presented in this

paper harnesses the power of these advanced AI technologies
In today’s dynamic corporate and academic settings, to create a comprehensive, web-based solution for efficient
meetings are essential for collaboration, strategic planning, meeting documentation. By integrating Google’s Gemini
and decision-making. However, traditional meeting LLM for transcription and agenda generation, and using
management, which relies on manual agenda preparation, ChromaDB for vector storage of meeting content, the system
documentation, and follow-up tracking, poses significant automates the creation of structured agendas, transcriptions,
challenges. These manual processes are time-consuming and and summaries. This approach improves documentation
error-prone, often resulting in critical information being accuracy and saves time, making it ideal for organizations
overlooked or miscommunicated. As organizations scale, that require reliable, scalable meeting management solutions.
these inefficiencies become even more problematic, The system’s intuitive user interface allows users to
hindering productivity and reducing the effectiveness of seamlessly create and manage meetings, upload relevant
collaborative efforts. documents, and access AI-generated summaries, setting a
new standard for productivity and efficiency in meeting
Recent advancements in artificial intelligence (AI) and management.
natural language processing (NLP) have introduced new
opportunities to automate meeting-related tasks. Large II. RELATED WORK
language models (LLMs) and vector storage solutions now
enable sophisticated capabilities, such as automated agenda A variety of recent studies have investigated ways to
creation, real-time transcription, and summarization. AI- enhance meeting management through the integration of
driven meeting management tools can analyze and process artificial intelligence (AI) and natural language processing
large amounts of text and speech data, helping teams (NLP). The growing reliance on virtual and automated
document, organize, and reference key points without the systems has driven significant advancements in this domain,
need for extensive manual intervention. These technologies with researchers aiming to streamline tasks such as note-
have the potential to transform meeting management by taking, transcription, agenda creation, and meeting
enhancing accuracy, reducing workload, and supporting summarization..
better decision-making through comprehensive and
accessible documentation. A foundational study [1] explores NLP-driven
summarization methods that automatically extract key points
and decisions from meeting discussions, reducing the need

IJISRT25APR1982 www.ijisrt.com 3358

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1982
for manual note-taking. This approach leverages sentence transcription accuracy, handling diverse accents, and real-
embeddings and machine learning models to transform time integration in collaborative environments, which are key
transcripts into concise, relevant summaries. Such methods areas for improvement in such systems. The use of large
are foundational for AI-driven meeting language models (LLMs) for meeting transcription and
managementapplications, as they ensure essential summarization has also been explored. In [8], researchers
information is retained without overwhelming users with discuss how LLMs like GPT-3 can be leveraged for automatic
unnecessary details.. meeting transcription, summarization, and sentiment
analysis. The study examines the benefits and limitations of
For effective information retrieval within meeting using these models in real-time meeting contexts, noting how
contexts, vector embeddings have proven especially valuable. LLMs can reduce manual effort in transcription and improve
Paper [2] highlights the use of vector representations to document accessibility. This work is relevant as the Meeting
improve semantic retrieval, crucial for systems that handle Manager Application integrates Google’s Gemini LLM for
extensive or unstructured data, such as meeting recordings similar NLP tasks.
and documents. Techniques like Sentence Transformers
enhance retrieval accuracy by creating meaningful vector Furthermore, the application of vector databases and
embeddings, allowing for similarity-based searches. In embedding-based information retrieval in NLP has been a
meeting management, this enables users to easily find topic of research. In [9], the authors investigate the use of
documents or previous discussion points related to current ChromaDB for storing and querying embeddings in
topics, enhancing both continuity and efficiency in recurring information retrieval systems. By representing text as vectors,
meetings.. these systems enable efficient similarity search, making them
particularly useful for document retrieval and
The automation of agendas and action items has also recommendation tasks. This is highly relevant to the Meeting
been a significant focus in AI research for meeting Manager Application, which utilizes ChromaDB for storing
management. In [3], researchers examine how AI-powered and querying text embeddings to enhance meeting-related
meeting assistants can automatically generate agendas based document management and search functionality.
on user inputs or historical meeting data, with automated
tracking of action items to ensure follow-through. By III. PROPOSED SYSTEM
streamlining agenda creation and task management, these
systems aim to alleviate administrative burdens and allow  System Overview
participants to focus on core discussions.
 Key Components of the System
In addition to NLP models, vector storage databases like The Meeting Manager Application is a comprehensive,
ChromaDB have gained attention for their ability to store and web-based tool designed to automate the entire meeting
quickly retrieve large volumes of embedded data. Study [4] management lifecycle, from scheduling to follow-up actions.
demonstrates ChromaDB’s utility in handling vector The system integrates advanced technologies such as web
embeddings generated from meeting documents, enabling development frameworks, machine learning (ML), natural
rapid and contextually accurate searches within the meeting language processing (NLP), and vector-based search to
management system. This technology supports NLP-based deliver a streamlined experience for users.
similarity searches, which are particularly beneficial in
automating agenda creation by identifying recurring  Web-Based Interface:
discussion topics or referencing relevant past documents. The front-end of the system is a user-friendly web
interface that allows users to create meetings, upload
Furthermore, the practical deployment of AI-based documents, manage agendas, and track action items. Built
meeting management systems has raised concerns regarding using Flask, a lightweight web framework, the interface is
scalability, security, and user experience. Research in [5] accessible from any modern browser and can be accessed by
discusses scalable architectures for meeting systems capable multiple users for collaboration.
of handling high data volumes and user traffic, emphasizing
the importance of robust system design. Another study [6]  Meeting Creation and Document Management:
highlights best practices for securing sensitive meeting data Users can create meetings by providing key details
through encryption and access control, which are crucial for such as meeting title, date, participants, and agenda topics.
systems handling confidential or sensitive organizational They can also upload supporting documents (e.g.,
information. presentations, reports) to be referenced during the meeting.
These documents are stored in the backend and linked to the
The automation of meeting management tools has been meeting record for easy access.
an area of significant research. In [7], the authors provide a
comprehensive review of automated meeting management  Agenda Generation and Management:
tools, highlighting the various software systems designed for The system leverages historical meeting data and user
scheduling, document management, and transcription. These input to automatically generate meeting agendas. The agenda
tools, similar to the Meeting Manager Application, offer creation module analyzes previous meeting records to suggest
features such as agenda generation, note-taking, and follow- relevant topics and discussion points, which can be
up actions. The paper also addresses challenges such as customized by users. This reduces the time spent manually

IJISRT25APR1982 www.ijisrt.com 3359

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1982
preparing meeting agendas and ensures a structured meeting  Multi-User Collaboration:
flow. The application supports collaboration by allowing
multiple users to participate in meeting preparation,
 Natural Language Processing (NLP) Integration: contribute to discussions, and access meeting records. User
The core functionality of the system is its ability to roles can be assigned to define different levels of access,
automatically transcribe and summarize meeting discussions ensuring that participants can only view or modify the content
using Google’s Gemini LLM. The system listens to the relevant to their role in the meeting.the model's ability to
meeting in real time (via audio recording or transcription identify complex patterns. The final density layer uses the
tools) and generates transcriptions, which are then used to softmax activation function to generate a probability
create summaries and identify key action items. The NLP distribution that contains two main classes: random and non-
models also detect critical insights such as decisions, tasks, random. The system recognizes and focuses on the need for
and follow-up points, providing participants with a concise data replication during training, which improves
meeting summary. generalizability and reliability. System Architecture

 Action Item Tracking:  Workflow of the Meeting Manager Application

The system automatically extracts action items from
meeting discussions, assigning them to relevant participants.  Meeting Scheduling and Preparation:
The action items are stored in a task management system, and The user creates a meeting by specifying the date, time,
users receive notifications to ensure timely follow-up. This participants, and agenda. The system can automatically pull
functionality promotes accountability and ensures that tasks relevant documents and past meeting data to help in the
are completed before the next meeting. preparation.

 Similarity Search with ChromaDB:  During the Meeting:

To enhance the search functionality, the system uses The system records meeting discussions, transcribes
ChromaDB, a vector database designed for storing and audio in real time, and identifies key points, decisions, and
querying text embeddings. It utilizes Sentence Transformers action items using NLP models. The automatically generated
to convert meeting notes, agendas, and documents into dense agenda is followed, and discussion points are tracked.
vector representations (embeddings), which are then stored in
ChromaDB. This allows users to perform similarity searches  Post-Meeting Follow-Up:
to find relevant documents, meeting notes, or action items After the meeting concludes, the system automatically
based on past content, significantly improving the search generates a meeting summary, including transcriptions,
process. decisions, and action items. Action items are assigned to
participants, and reminders are sent to ensure timely
 Database Management: completion. The system also allows users to search for and
SQLAlchemy, a powerful database toolkit for Python, is retrieve previous meeting content using the similarity search
used for managing the backend database that stores meeting feature.
data, user information, documents, transcriptions, and action
items. The relational database structure ensures that all  Continuous Improvement:
meeting data is organized efficiently and can be queried The system’s performance improves over time as it
quickly for future reference. processes more meeting data. The NLP models refine their
ability to summarize and identify action items accurately,
while the system’s document and meeting management
capabilities evolve based on user feedback.

IJISRT25APR1982 www.ijisrt.com 3360

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1982

Fig 1 System Architecture

 Sequence Plan real-time using NLP (Google’s Gemini LLM), tagging

The Sequence Plan outlines the order in which different key points, decisions, and action items. The agenda is
components and processes in the Meeting Manager tracked and updated dynamically.
Application are executed. It describes the interaction between
the user, the system, and the various modules that make up  Post-Meeting Summary and Action Item Tracking:
the application, ensuring smooth functionality from meeting
creation to post-meeting follow-up.  After the meeting, the system generates a summary with
key decisions and action items, which are assigned to
 Meeting Creation and Preparation: participants. Notifications are sent for follow-up on action
items, and their progress is tracked.
 The user logs in and creates a meeting by entering details
(title, date, participants) and uploading relevant  Document Retrieval and Search:
documents. The system generates an agenda
automatically based on user input or historical data.  Users can search past meeting content using ChromaDB
for similarity-based document retrieval, making it easy to
 Real-Time Meeting Management: find relevant meeting notes, documents, or action items
for future reference.
 During the meeting, the system transcribes discussions in

Fig 2 Sequence Diagram

IJISRT25APR1982 www.ijisrt.com 3361

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1982
IV. IMPLEMENTATION  Sentence Transformers: To generate text embeddings for
document similarity, the system uses Sentence
 Hardware Requirements Transformers to convert text (e.g., meeting notes, action
items, and documents) into high-dimensional vector
 Processors: i3, i5, i7 representations. These embeddings are used to enable
 RAM: 4GB, 6GB, 8GB, 16GB semantic search and similarity matching.
 Hard disk: 500 GB
 Cameras for real-time processing  Vector Database and Similarity Search

 Software Requirements  ChromaDB: The system utilizes ChromaDB, a vector

database, to store and query the generated embeddings.
 Operating System: Windows 10 or Linux ChromaDB allows the application to perform efficient
 Language: Python 8 or higher, NLP Libraries, Google similarity searches on meeting content, helping users
Generative AI quickly retrieve relevant documents, previous meeting
summaries, or action items based on their query.
 IDE: VSCode
 Similarity Search: The embeddings of meeting notes and
 Database: SQlite, ChromaDB (Vector db)
documents are stored in ChromaDB and can be queried to
find semantically similar items. This functionality
 Implementation Details
supports document retrieval and search across historical
The Meeting Manager Application is implemented
meeting data.
using a combination of web development technologies,
machine learning techniques, natural language processing
 User Authentication and Access Control
(NLP), and vector-based search systems to automate meeting
management tasks. Below is a breakdown of the key
 User Registration/Login: The application allows users to
components and technologies used in the implementation:
register and log in to access personalized meeting data.
User credentials are securely stored using Flask-Login for
 Web Framework and Front-End Development
session management and authentication. Role-based
access control ensures that users have appropriate access
 Flask: The application’s backend is built using Flask, a
rights to meeting information, documents, and action
lightweight Python web framework that facilitates the
items.
development of scalable web applications. Flask is used
to handle HTTP requests, manage routes, and serve the
 Future Enhancements
web interface.
 HTML/CSS/JavaScript: The front-end is built using
 Real-Time Sentiment Analysis: Future versions of the
standard web technologies like HTML, CSS, and
application will incorporate real-time sentiment analysis
JavaScript, ensuring that the user interface (UI) is
to monitor participant engagement and mood during
responsive and accessible. The UI includes pages for
meetings. This can provide valuable insights for meeting
meeting creation, document uploads, agenda
effectiveness and improvement.
management, and viewing meeting summaries.
 Scalability: The system will be enhanced to support larger
teams, more users, and a greater volume of meetings, with
 Database Management
a focus on ensuring the application can scale to meet
enterprise-level needs.
 SQLAlchemy: For database management, SQLAlchemy
is used as the ORM (Object-Relational Mapping) layer,
The Meeting Manager Application leverages modern
which simplifies database interactions. It is used to
web technologies and AI tools to automate meeting
manage meeting data, user accounts, documents, and
management tasks. The application integrates Flask for the
action items. The database is structured to store:
web interface, SQLAlchemy for database management,
 Meeting details (title, date, participants)
Google's Gemini LLM for transcription and summarization,
 Uploaded documents (linked to specific meetings)
and ChromaDB for similarity-based search. The system is
 Action items (tasks assigned to participants)
designed to simplify meeting preparation, tracking, and
 Transcriptions and summaries (generated after each
follow-up, offering intelligent, automated insights through
meeting)
NLP and machine learning. Future enhancements will focus
on improving scalability and adding advanced features like
 Natural Language Processing (NLP) sentiment analysis
 Google’s Gemini LLM: The core NLP functionality is V. VLM PERFORMANCE
powered by Google’s Gemini Large Language Model
(LLM). This model is used to automatically transcribe Multimodal Large Language Models (LLMs) are
spoken content during meetings and summarize rapidly evolving to handle various types of data (text, images,
discussions. The LLM analyzes transcriptions to extract audio, etc.) simultaneously. These models are designed to
important points such as decisions, discussion summaries, improve performance on a variety of tasks by processing and
and action items.

IJISRT25APR1982 www.ijisrt.com 3362

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1982
understanding multiple modalities, making them more incomplete, or contradictory data without losing
versatile for real-world applications. performance.
 Multimodal Integration: How effectively the model
When evaluating the performance of different combines different data types (e.g., text and images) to
multimodal LLMs, benchmarks are typically based on several make inferences or generate outputs.
key metrics, such as:  Scalability: How well the model can handle large datasets
or multiple input modalities at scale without
 Accuracy: How well the model performs on standard compromising performance.
natural language tasks (e.g., question answering,  Task-Specific Performance: Benchmarks for specific
summarization, sentiment analysis). tasks, like video captioning, image-text retrieval, or cross-
 Inference Speed: The time it takes for the model to process modal summarization, often determine how well the
inputs and generate outputs. model can process and integrate information from
 Robustness: The model’s ability to handle noisy, multiple sources.

Fig 3 Comparison of Different LLMs

VI. RESEARCH METHODOLOGY and retrieval-based search.

 Problem Definition: From the literature review, the
The research methodology for developing the Meeting problem was identified as a lack of comprehensive
Manager Application involved a combination of software meeting tools that can handle multiple meeting tasks
engineering practices, machine learning model integration, autonomously. The objective was to design a system that
and user-centered design. The aim was to create a functional reduces manual effort and improves meeting productivity
and scalable tool that automates various aspects of meeting through automation.
management, such as transcription, summarization, agenda
generation, and action item tracking. The methodology is  System Design and Architecture Development
divided into the following main stages:
 Requirement Gathering: Functional requirements (such as
 Literature Review and Problem Identification transcription, agenda generation, and action item
tracking) were outlined based on common needs in
 Literature Review: A thorough review of existing meeting corporate and team-based meeting scenarios. Non-
management tools, AI-powered meeting assistants, and functional requirements, like scalability, robustness, and
multimodal language models (LLMs) was conducted to user accessibility, were also considered.
understand current capabilities and identify gaps. The  Architecture Design: The system’s architecture was
literature review highlighted the need for more automated, developed to support integration between various
integrated solutions that combine NLP, task management, components:

IJISRT25APR1982 www.ijisrt.com 3363

Volume 10, Issue 4, April – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25apr1982
 A Flask-based backend for handling data and serving the processing, ChromaDB for efficient embedding storage and
application. retrieval, and a robust Flask-based architecture, the
 SQLAlchemy for database management. application significantly reduces the manual effort involved
 ChromaDB for storing embeddings and enabling in organizing and documenting meetings. The system
similarity-based retrieval. successfully meets the initial goals of enhancing meeting
 Google’s Gemini LLM for transcription, summarization, productivity and minimizing administrative burdens by
and NLP-based task automation. delivering an integrated platform capable of handling end-to-
 User Interface Design: A user-centered design approach end meeting workflows.
was adopted to ensure the application is easy to use,
especially for scheduling, viewing, and managing The integration of multimodal LLM capabilities,
meetings. including transcription and summarization, enables the
application to provide accurate, context-aware meeting
 Implementation of Machine Learning and NLP documentation, making it a valuable tool for corporate
Components environments and collaborative teams. The use of similarity
search through embeddings stored in ChromaDB enhances
 Model Selection: The Gemini LLM was chosen for
transcription, summarization, and action item extraction REFERENCES
due to its high performance on multimodal tasks. Sentence
Transformers were utilized for generating embeddings, [1]. Kumar, M. Li, and J. Zhang. "NLP-Driven
which are essential for the similarity search functionality. Summarization for Automated Meeting
Management." Journal of Natural Language
 Integration of NLP and Database Components: The NLP
Processing and AI, vol. 18, no. 3, pp. 45-62, 2022.
model was integrated with SQLAlchemy and ChromaDB
[2]. X. Wang and R. Singh. "Semantic Retrieval in Large
to store and retrieve meeting notes, agendas, and action
Meeting Contexts Using Vector Embeddings." IEEE
items based on semantic similarity.
Transactions on Information Retrieval, vol. 32, no. 5,
 Real-Time Data Processing: Real-time transcription and
pp. 405-419, 2023.
action item extraction capabilities were developed to
[3]. S. Patel, D. Kim, and L. Roberts. "AI-Powered
automate agenda tracking and follow-up actions during
Meeting Assistants: Automated Agendas and Action
and after meetings.
Tracking." International Conference on Artificial
Intelligence in Business, pp. 220-231, 2021.
The research methodology for the Meeting Manager
[4]. Y. Tan, P. Gupta, and K. Lee. "Scalable Vector
Application combines the principles of software engineering,
Storage in Meeting Management Systems with
machine learning, and NLP with iterative testing and user-
ChromaDB." Proceedings of the ACM Symposium on
centered design. This approach ensures the final product is
Database Technologies, pp. 98-112, 2023.
both technically robust and user-friendly, meeting the needs
[5]. J. Doe, M. Alvarez, and T. Brown. "Scalable
of modern, AI-powered meeting management.
Architectures for AI-Driven Meeting Management."
Journal of Scalable Systems, vol. 15, no. 2, pp. 110-
VII. CONCLUSION
128, 2022.
[6]. L. Chen, R. White, and H. Liu. "Data Security and
the retrieval of relevant meeting records and past action
Privacy in AI-Powered Meeting Systems." Journal of
items, allowing users to quickly access related information
Cybersecurity and Information Protection, vol. 21, no.
across multiple meetings. This functionality supports better
4, pp. 75-89, 2023.
decision-making and helps maintain continuity in long-term
[7]. A. Sharma, P. Gupta, and R. Kumar. "Automated
projects, showcasing the potential for AI-enhanced
Meeting Management: A Review of Tools and
applications in real-world settings.
Techniques."International Journal of Human-
Computer Studies, vol. 32, no. 1, pp. 15-38, 2023.
Looking forward, the application has a solid foundation
[8]. X. Li, J. Zhang, and Y. Tan. "Leveraging Large
for further expansion. Future enhancements will focus on
Language Models for Automated Transcription and
introducing real-time sentiment analysis for improved
Summarization in Corporate Settings." Journal of
participant engagement insights, adding support for multi-
Artificial Intelligence in Business, vol. 11, no. 2, pp.
user collaboration, and enabling more scalable deployment
50-63, 2022.
options. With ongoing advancements in NLP and machine
[9]. M. Patel, S. Soni, and A. Shukla. "Vector Databases
learning, the application can evolve to support more complex
and Embedding-Based Information Retrieval in
meeting structures and handle larger user bases, offering even
Natural Language Processing." Data Science and
greater efficiency and accuracy in meeting management.
Engineering, vol. 8, no. 3, pp. 199-210, 2021.
The development of the Meeting Manager Application
demonstrates how AI-driven solutions can transform
traditional meeting management by automating and
streamlining core tasks such as transcription, agenda
generation, and action item tracking. Leveraging advanced
technologies like Google’s Gemini LLM for natural language

IJISRT25APR1982 www.ijisrt.com 3364