01
A single extraction pipeline
Problem
Each tool read uploaded files its own way, so the same document produced different results.
How we solved it
All features now share one extractor with stable indexes for text and embedded images.
Case study · 19Content Operations
SimplexED Suite
Nine content tools for corporate training teams, from slides to voiceover to video, working from one source file and one AI layer.
In active development
- Client
- SimplexED Suite
- Industry
- Corporate L&D · Content Operations
- Platform
- Web
- Timeline
- 14 months, ongoing
- Outcome
- Nine authoring tools working from a single project library
- 9Production tools in one workspace
- 1Shared library and AI layer
- 14 moOf development, ongoing
01
Overview
SimplexED Suite lets learning and enablement teams take a single source document and publish it as a guide, a branded deck, a recorded walkthrough or a voiced video. We delivered nine production tools inside one workspace, and every tool reads from the same project library.
02
The challenge
Training content was being rebuilt by hand for every format, in separate apps and from separate copies of the source. Versions drifted apart until the video no longer matched the guide. Off-the-shelf AI slide makers made things worse, producing decks that looked nothing like the company's own template and read the same no matter who the audience was.
03
Our approach
Our guiding decision was that a file should mean the same thing everywhere. One extraction service feeds every tool, generated slides are placed into the customer's real template, and the heavy media work happens in the user's browser.
04
How we built it
4 steps, from architecture to launch.
- 01
One way to read every file
A shared service pulls text and images from DOCX, PDF and PPTX with stable references that every AI feature reuses.
- 02
Brand templates, analyzed
Our template analyzer, revised through twelve versions, classifies real slide designs so generated content lands in the right boxes.
- 03
Video without servers
Recording, editing and MP4 export run on WebCodecs, and the preview shares the export's rendering code.
- 04
Queued generation
BullMQ workers on Redis, managed by PM2, build branded decks in the background and finish cleanly during restarts.
05
Problem → How we solved itTechnical challenges
02
Generated content in the customer's own template
Problem
Corporate templates position text and imagery precisely, and generic generators ignored them.
How we solved it
The analyzer strips placeholder text and fits new content into the actual layouts, keeping masters, themes and fonts.
03
Audience before prompt
Problem
A single prompt box gave executives step-by-step instructions and gave operators vague overviews.
How we solved it
Deck generation starts by asking for the angle (conceptual, executive or procedural) and enforces it across every slide.
06
6 screensScreens
07
6 featuresWhat we shipped
- 01
Decks in your template
Smart PowerPoint builds slides from a prompt or document and fits them into an uploaded .pptx.
- 02
Documents with AI help
A browser-based DOCX editor with offline-first storage, autosave and paragraph-level rewrite, tone and translation.
- 03
Scripts and voices
Narration scripts from any source, voiced through ElevenLabs with speed control and waveform preview.
- 04
Capture and edit video
Screen, window or tab recording plus a multi-track editor with captions, all exported as MP4 locally.
- 05
Images
A Fabric.js editor for cropping, filters, text and shapes, choosing PNG or JPEG automatically.
- 06
Shero assistant
Answers product questions from a Pinecone knowledge base, streams replies and hands off to a person.
08
Tech stack
- React
- Single-page authoring workspace
- Node.js + Express + MongoDB
- TypeScript REST API and data
- Gemini Flash
- Multimodal AI and document RAG
- WebCodecs
- In-browser MP4 video engine
- BullMQ + Redis
- Background deck generation
- ElevenLabs
- Voiceover generation
- Pinecone
- Assistant knowledge base
Tell us what you are shipping.
Start a Project →