Agentic AI PowerPoint Builder: A Multi-Agent System for Automated PPTX Presentation Generation

Abstract

Creating PowerPoint presentations is a time-consuming task requiring structured planning, content writing, and visual design. This project introduces the Agentic AI PowerPoint Builder, a multi-agent AI system that automatically generates complete professional presentations (.pptx) from simple user inputs such as topic, number of slides, content depth, font style, and image inclusion.

The system is built using a LangGraph-based multi-agent workflow, where each agent specializes in a distinct responsibility: slide structuring, content drafting, image sourcing, and final PowerPoint assembly. The final output is an exportable, downloadable PowerPoint file generated end-to-end with minimal user effort.

This project was developed to fulfill the requirements of the ReadyTensor Agentic AI Developer Certification – Module 2, demonstrating a tool-integrated multi-agent system with real-world usability.

1. Motivation

Presentation creation is a common requirement in education, business, and technical domains. However, manual slide building involves several repetitive steps:

• Designing a slide outline

• Writing concise and relevant slide content

• Finding supporting images

• Formatting slides consistently

• Exporting into PPTX format

While large language models can generate text, generating a full structured presentation with visuals requires multiple coordinated reasoning steps. This makes the task ideal for a multi-agent AI architecture, where agents collaborate like a team.

2. Problem Statement

A single-agent AI response is insufficient for producing full PowerPoint presentations because presentation building requires:

• Planning across multiple slides

• Maintaining structure and flow

• Controlling formatting and style

• Retrieving relevant images

• Generating an actual .pptx file

Thus, the goal of this project is to build an Agentic AI system that coordinates multiple specialized agents to automatically generate complete PowerPoint presentations.

3. System Overview

The Agentic AI PowerPoint Builder takes user inputs:

• Presentation Topic

• Number of Slides

• Font Style (Times New Roman, Arial, Calibri, etc.)

• Content Type (Minimal / Concise / Detailed)

• Image Inclusion Preference

The system then produces:

1. Slide-wise structured presentation
2. Relevant content and visuals
3. A fully formatted PPTX export ready for download

4. Multi-Agent Architecture

The solution is implemented as a true multi-agent workflow using LangGraph orchestration, where each agent has a dedicated role.

Agents Included

Agent 1: Slide Planner Agent

Breaks the topic into an ordered slide outline

Determines slide titles and presentation flow

Agent 2: Content Writer Agent

Generates bullet-point content for each slide

Adjusts depth based on user preference (minimal, concise, detailed)

Agent 3: Visual & Image Agent

Produces relevant image keywords per slide

Fetches visuals through the Unsplash API

Ensures each slide includes supporting imagery

Agent 4: PPT Builder Agent

Assembles all slides into a PowerPoint file using python-pptx

Applies consistent font formatting

Exports the final downloadable .pptx presentation

Figure 1: LangGraph-orchestrated multi-agent architecture for automated PPT generation.

5. Tool Integration (Module-2 Requirement)

To meet ReadyTensor Module-2 requirements, the project integrates multiple external tools beyond LLM generation:

• ChatGroq (Llama-3 70B) – High-quality slide content generation

• LangGraph Framework – Multi-agent orchestration and state coordination

• Unsplash API – Retrieval of royalty-free relevant images

• python-pptx – Programmatic PowerPoint file creation

• Streamlit UI – User interface for inputs and PPT download

This satisfies the requirement of using 3+ integrated tools within a multi-agent pipeline.

6. Workflow Execution

The system operates sequentially through agent collaboration:

User Input
→ Slide Planner Agent
→ Content Writer Agent
→ Visual & Image Agent
→ PPT Builder Agent
→ Final PPTX Export

Each agent passes structured outputs through LangGraph’s shared state, ensuring modular coordination and clean division of responsibility.

7. Demo Interface and Example Run

The system provides a simple Streamlit-based interface where users can enter:

• Presentation topic
• Number of slides
• Font style
• Content depth
• Image inclusion preference

Once submitted, the multi-agent pipeline generates a complete downloadable PPTX file.

Figure 2 below shows the working UI used for presentation generation.

Figure 2: Streamlit interface for Agentic AI PowerPoint Builder.

8. Key Features

• Fully automated PowerPoint generation

• Multi-agent collaboration with clear role separation

• Adjustable content depth (minimal/concise/detailed)

• Font customization across slides

• Automatic relevant image insertion

• Exportable .pptx file download

• Extendable architecture for future improvements

9. Future Enhancements

Potential improvements include:

• Theme templates (corporate, academic, creative)

• Human-in-the-loop slide editing

• Support for charts and tables

• Slide quality evaluation metrics

• Deployment as a full SaaS PPT generator

Conclusion

The Agentic AI PowerPoint Builder demonstrates how multi-agent AI systems can solve complex real-world tasks beyond text generation. By coordinating specialized agents for slide planning, writing, visual sourcing, and file construction, the system delivers complete professional PowerPoint presentations automatically.

This project successfully fulfills ReadyTensor Module-2 objectives by implementing:

• 3+ collaborating agents
• 3+ integrated tools
• LangGraph-based orchestration
• Real-world end-to-end usability