Applied AI / LLM
AI Deck Studio
A document-to-deck studio: paste a document or a prompt, let an LLM draft the outline and slides, then narrate every slide in a cloned voice. It turns a rough brief into a presentable deck — and an auto-narrated walkthrough video — in one pass.
What it does
- Draft with AI — turn a document or a short prompt into a structured outline and slide content.
- Voice cloning — upload a short reference clip to clone a voice, preview it, and set it as the deck narrator.
- Per-slide narration — generate narration audio per slide, with per-slide voice overrides and a reset-to-deck-voice default.
- Export — export to
.pptx, re-import a.pptxfor editing, and render a narrated walkthrough video.
The voice
Narration runs on an XTTS v2 model I fine-tuned for a target speaker. Cloning takes a short reference clip; the studio caches voice embeddings so repeat narration stays fast. Every word in the video above is synthetic.
How it was built
Built in about a week with an agentic-coding workflow — AI coding agents driving the implementation under close review — across the generation, slide-editing, voice, and export layers. The fine-tuned XTTS v2 model is the piece that makes the narration sound like a specific person rather than a stock TTS voice.