AI Automation: Deploying an Autonomous Agent Workflow for Slide Creation
Automation Stack & Architecture
- Agent Framework: Multi-agent system involving Content Planning Agents (LLM-based such as ChatGPT or Claude) and Asset Retrieval/Assembly Agents.
- Integration Layers: GitHub (Open-Slide project), Search API integration (Yandex.Images), and File Export module (.pptx).
- Target Outcome: Rapid generation of structured, image-rich presentation files from a single natural language prompt.
Agent Roles & Tools Assignment
- Content Planner Agent
- Persona: Strategic content architect responsible for slide structure and copy drafting.
- Goal: Transform high-level topics into detailed outlines consisting enough reasonable number of slides.
- Tools assigned: Large Language Models (`ChatGPT`, `Claude`).
- Asset Integration Agent
- Persona: Visual researcher focused on thematic relevance.
- Goal: Identify and retrieve relevant imagery to complement the text instructions via external search APIs.
- Tools assigned:
Function №198 - Yandex Images Finder.
- Presentation Assembler Agent
- Persona: Layout engineer managing object placement/formatting in PPTX format.
- Goal: Convert generated code or structures into discrete PowerPoint objects (text, shapes, images) that remain independently editable.
- Tools assigned:
open-slide 2.0 / open source engine.
Step-by-step Workflow Orchestration
- User provides a natural language prompt such as
"Make a presentation about AI trends for 10 slides"which acts as the initial trigger. - The Content Planning agent processes this request using an LLM (
ChatGPT_or_Claude) to generate slide structure and content drafts. - Simultaneously or sequentially, the Asset Integration agent executes
get_relevant_images()by querying image databases like Yandex enough matching visual assets per topic description. - The system aggregates text data and retrieved imagery through the assembly toolset where any technical requirements (`size`, `positioning`) are applied via scriptable commands if certain tools allow manual adjustment post-generation.
- Final export protocol triggers the creation of a .pptx file containing separate object layers including wayfully formatted text/shapes before delivering the final output package.
Error Handling & Loop Prevention (Human-in-the-loop validation)
- Manual Refinement Gate: Instead of relying solely on autonomous generation, humans act as editors (
! DYOR (Do Your Own Research)