From hot news to post-ready content.

Content Agent is an end-to-end AI content system independently designed and developed by Yuxin for companies, agencies, and solo creators producing content across multiple platforms.

The short-form era turned every brand and creator into a multi-platform publisher. Content Agent is the tool I wanted: one system that sits on top of the entire workflow, from trend discovery and content ideation to generation, publishing, and performance feedback.

Most AI content tools have one trick: they turn anything into an AI-generated video. But a few seconds of generated footage cannot always explain a market move, regulatory shift, or tech announcement, and matching abstract news to convincing visuals is often a losing game.

Content Agent starts from the opposite premise.

For every trending topic it discovers, an LLM judge selects the optimal format: video when the topic itself is visual, such as a runway show or sports moment; an image with voiceover for narrative news; a multi-image slideshow with TTS narration for finance and tech; a carousel for list-style content; or a text post for a concise take.

Each format runs through its own dedicated pipeline. Runway and fal.ai models—including Seedance, Veo 3, Kling, and Hailuo—handle visual generation; Edge-TTS handles multilingual voiceover; FFmpeg assembles slideshows with Ken Burns motion; and DeepSeek drives ideation, copywriting, and format decisioning.

Every step stays human-in-the-loop. Users can edit scene prompts, rewrite narration, modify image prompts in raw JSON, or override the agent’s format choice before generation.

A SQLite-backed feedback loop closes the system. After publishing, users can log a post’s actual performance. DeepSeek then distills high-performing content into platform-specific style guides that feed back into future batches as few-shot context—allowing the system to become more tailored with every post shipped.

Yuxin independently designed and developed the complete system, from product concept and agent architecture to implementation and API integration, using OpenAI Codex, Claude Code, DeepSeek API, Runway, Kling, Seedance, Veo 3, and fal.ai.