Content Agent

Content Agent is the most ambitious project I've built. It's an autonomous AI agent that takes a scene description and produces production-ready video content — handling everything from storyboard planning to final export.

The agent has 74 tools organized across several domains: Neo4j queries for managing the knowledge graph of scenes, characters, and locations; FAL AI generation via Flux and Nano Banana Pro with model routing based on the task; video generation using Sora for dialogue-heavy scenes and Veo for motion-heavy scenes; storyboard management; video coherence for ensuring visual consistency across shots; and image manipulation for post-production.

The hardest challenge in AI video production isn't generating a single good shot — it's making 30 shots look like they belong in the same film. Content Agent solves this with a coherence system that maintains reference images for every character and environment, feeds them as context to each generation, and flags inconsistencies for re-generation.

Most AI video tools generate first and edit later. Content Agent flips this: it plans the edit first, then generates footage to fit the plan. The agent creates a storyboard with specific shot types, durations, and transitions, then generates images and video to match each panel. This is more efficient and produces better results — every shot has a purpose in the narrative.

The Neo4j knowledge graph is the backbone. Every character, location, prop, and relationship is stored as nodes and edges.