From research to deck: AshnaAI’s agentic presentation pipeline
AshnaAI does not generate a PowerPoint from one prompt. A create orchestrator runs request analysis, research, planning, image generation, and code. Updates use a separate orchestrator so an existing deck is not rebuilt from scratch.
Method: Static analysis of the create and update presentation orchestrators, including named pipeline stages and the research sub-steps context, search, and compile.
Key findings
- Create runs five stages: request analysis, research, plan, images, then style-specific HTML/CSS/JS streamed to the client.
- Research itself has three sub-steps—context, search, and compile—before a slide plan is written.
- Updating an existing deck uses a specialised orchestrator and does not rerun the full create pipeline.
Why a pipeline instead of one prompt
A slide deck fails in specific ways: weak research, inconsistent style, missing images, or a plan that does not match the requested length. AshnaAI splits those failures across named stages so the product can show progress and recover a stage instead of discarding the whole job.
The create orchestrator is the coordinator. It receives the user request and runs sub-agents in sequence. A pipeline tracker emits stage state to the client while work is in flight.
Named stages
Stage 0 analyses the request: content type, slide count, and planning inputs. Stage 1 researches from web and user context, with sub-steps for context, search, and compile. Stage 2 writes slide plans and selects style. Stage 3 generates style-matching images in parallel. Stage 4 generates style-specific HTML, CSS, and JS and streams it to the client.
Those names are the product contract. They are what a user sees moving, and what this article means by “agentic presentation pipeline.”
Create versus update
Editing a finished deck is a different job from creating one. The update orchestrator does not rerun request analysis through code generation as if the deck did not exist. That split avoids the common failure where a small change regenerates every slide.
Teams that need a how-to should start with the PowerPoint blog. Teams that need a work-use case should use the AshnaAI for work solution page.
Limitations
The pipeline produces presentations and infographics. It does not ship a public website, a web app, or a design system for your product UI.
This note reports architecture, not a timed bake-off against Copilot or Gamma. Those comparisons belong on commercial pages, with this article as the method behind the AshnaAI side.
Frequently asked questions
- Can AshnaAI generate a PowerPoint from a topic?
- Yes. The product runs an agentic pipeline that researches the topic, plans slides, generates images, and emits style-specific slide code. It is not a single one-shot completion.
- What are the stages of the presentation pipeline?
- Request analysis, research, plan, images, and code. Research is further split into context, search, and compile.
- Does an edit rebuild the whole deck?
- No. Updates go through a specialised update orchestrator so an existing presentation is revised instead of regenerated from zero.
- Does the pipeline use my files or only the web?
- Research gathers web results and user context. If you attach files, grounding happens through the file-embedding path described in the sibling research note.
- Is this the same as exporting a ChatGPT answer to slides?
- No. ChatGPT can draft outline text. AshnaAI’s pipeline tracks stages, plans individual slides, generates matching images, and streams renderable slide code.
- Can AshnaAI design a marketing website instead of a deck?
- No. Website and web-app creation are out of scope. The artifact path here is presentations and infographics.
Related
Try this in AshnaAI. Create a free account.