# Turn PowerPoint Templates Into AI Slide Builders (No Code Tutorial)

Source: https://www.youtube.com/watch?v=OtE4j5dFYxY
Recap page: https://rapidrecap.app/video/OtE4j5dFYxY
Generated: 2026-01-23T22:07:05.598+00:00

---
## Quick Overview

The tutorial demonstrates how to convert existing PowerPoint or Keynote slide templates into functional Relevance AI slide templates by using the built-in Slide Builder agent to recreate the design based on uploaded images or PDFs and then defining dynamic content fields using double square brackets for AI agents to populate automatically.

**Key Points:**
- The process involves using the Slide Builder agent in Relevance Chat to generate a brand kit from images (minimum three required) to capture colors, text styles, and logos from an existing template.
- Users upload their existing deck (e.g., PDF) and instruct Slide Builder to recreate it, noting that initial recreation requires manual iteration, such as fixing backgrounds and alignments, which can take 15 minutes to half an hour for an initial version.
- Templatable variables within the slides must use double square brackets (e.g., [[company name]]) because double curly brackets are reserved for variables within the Relevance platform.
- A custom research agent, named 'Taylor' in the demo, is integrated via a 'Pitch Builder' workforce to research account details (like Nike or Walmart) and populate the pre-defined fields in the AI-generated slide deck.
- The AI-first approach means manual edits are disabled; users must use labeling changes and annotations within the chat interface to refine the AI's output, emphasizing delegation skills.
- The final output can be shared as a link maintaining web app functionality like animations or exported as PDF, JSON, or PowerPoint, with sharing the link being the preferred method to retain interactive elements like pricing calculators.
- The session confirmed that agents can be embedded into slides, citing examples like 3D games and inbuilt quizzes for educational sectors, though image generation prompts are recommended to be handled by prompting the image generator agent separately in chat.

**Context:** The session, hosted by Alex from Relevance's community team and featuring solution engineer Tus, served as a live tutorial on leveraging Relevance AI's Slide Builder capability to automate presentation creation. The core objective is to take established company templates from platforms like PowerPoint or Keynote and transform them into dynamic, agent-ready templates within Relevance AI, ensuring brand consistency while automating content population via integrated AI research agents.

## Detailed Analysis

Tus walks through converting an existing slide deck into a Relevance Slide Template to automate pitch deck creation using AI agents. The workflow starts by defining a brand kit using screenshots from the source deck to capture color schemes and typography, followed by uploading the deck itself for the Slide Builder agent to attempt a recreation. This initial recreation necessitates manual iteration within the chat interface to correct alignments, backgrounds, and ensure brand compliance; users must label changes rather than editing directly, adhering to Relevance's AI-first philosophy where delegation skills are key. Crucially, dynamic content placeholders in the template must be marked with double square brackets, like [[company name]], which serve as explicit markers for the connected research agent (like 'Taylor') to target during population. Taylor, the research agent, is shown utilizing tools like Google Search for external data (e.g., company goals) and internal knowledge bases for vetted use cases, structuring its output into specific fields required by the slide template variables. Finally, these components are wired together in a 'Pitch Builder' workforce, where Taylor passes its research findings, formatted via a custom message prompt detailing which template fields to update, directly to the Slide Builder, automating the population of multiple slides for target companies like Walmart or Nike, thereby saving significant time on formatting and research integration.

### Template Conversion Process

- Uploading PDF/PPT template for AI recreation
- Iteratively fixing alignments and backgrounds using the chat interface
- Defining dynamic fields using double square brackets [[variable]] for agent integration

### Brand Kit Generation

- Slide Builder generates brand guidelines from images covering colors, text styles, and logos
- Users can input dark and light logos and define brand tones/voice
- The brand kit ensures consistency across automated slides

### Agent Integration Setup

- A research agent ('Taylor') is built to take a company name as input and research necessary data
- Agent uses tools like Google Search for external data and internal knowledge bases for vetted use cases
- Agent output fields must explicitly match template variables for successful population

### Workforce Automation

- The 'Pitch Builder' workforce connects the research agent (Taylor) to the Slide Builder agent
- A custom message prompt in the AI connection provides strict instructions to Taylor on which fields to populate in the template
- This setup enforces guardrails, ensuring the agent only updates specified placeholders based on research

### User Interaction & Iteration

- Manual edits are intentionally disabled; users must communicate desired changes via labeling changes and annotations in the chat
- This requires users to improve delegation skills, treating the AI as a delegate that needs precise instruction

### Output and Sharing Options

- Final slides can be exported as PDF (with editable text or image-only), JSON, or PowerPoint
- Sharing the direct web link is preferred as it retains interactive web app functionality like animations and ROI calculators that static downloads lose