Automated video tools often match isolated keywords and produce irrelevant footage, empty scenes, or generic stock that does not support the story.
Case Study 02 — AI Media
MAIRA Studio
Local-first AI video production platform combining narrative intelligence, semantic asset selection, provider orchestration, voice generation, and media processing.
SCRIPT → FILM
Problem
Solution
The workflow first understands the full narrative, then builds scenes and determines visual intent.
Full Script→Narrative Analysis→Scene Plan→Asset Intelligence→Render
- Entity, timeline, mood, event, and scene-purpose analysis.
- Fallback hierarchy from exact archival match to contextual assets and generated visuals.
- Automatic retry when a scene lacks usable media.
- Local-first FFmpeg/FFprobe processing to keep heavy rendering on customer hardware.
Engineering
Provider APIs are treated as replaceable capabilities behind an authoritative product workflow. Runtime hardening covers malformed AI output, missing assets, provider failures, subtitle leakage, codec fallback, and render recovery.
The product is an editorial system with AI inside it — not an AI call wrapped in a video UI.
My Role
Product workflow, AI analysis stages, scene planning rules, asset fallback strategy, local-vs-cloud boundaries, provider orchestration, runtime hardening, and production testing.