Inspiration
Trading is full of information, but too much information can make it difficult for traders to know what actually matters. We wanted to build a system that combines AI research, creative storytelling, and video production to automatically create better trading advertisements.
What it does
CrowdWisdom AI Studio is an autonomous multi-agent system that researches competitor advertisements, identifies trader pain points, creates customer profiles, and generates creative advertising concepts.
It automatically creates scripts, storyboards, voiceovers, sound design, and vertical 9:16 video advertisements. It also performs quality and financial compliance checks before the final video is delivered.
How we built it
We built CrowdWisdom AI Studio using a Hermes-based multi-agent architecture. Each AI agent has a specific role, including competitor research, marketing analysis, pain-point research, customer profiling, creative direction, scripting, storyboarding, video generation, and quality assurance.
The system is built with Python, Pydantic, FFmpeg, OpenMontage/Hyperframes integrations, and procedural video generation. We also integrated research and LLM APIs while adding caching and fallback systems to keep the pipeline reliable.
Challenges we ran into
The biggest challenge was coordinating multiple AI agents and making sure every stage produced reliable and structured results.
We also had to handle API failures, missing API keys, video rendering, data validation, and financial advertising compliance. To solve these problems, we added structured schemas, deterministic caching, automated QA checks, and a procedural video-generation fallback.
Accomplishments that we're proud of
We built a complete autonomous pipeline that can go from competitor research to a finished advertising video.
The system can generate 1080x1920 vertical advertisements with voiceover, sound design, animated market visuals, and automated compliance validation.
We are also proud of the zero-API demo mode, structured output artifacts, caching system, and automated test suite with all 12 core tests passing.
What we learned
We learned that building a reliable multi-agent system requires more than good AI prompts. Agents need clear responsibilities, structured communication, validation, error handling, caching, and reliable fallbacks.
We also learned how important research and source traceability are when using AI to create financial content.
What's next for CrowdWisdom AI Studio
Next, we want to integrate advanced generative video APIs to create more realistic advertisements.
We also plan to add an A/B testing feedback loop using real advertising performance data so the system can learn which creative concepts perform better.
Another planned improvement is multi-character voice dialogue to make the generated advertisements more cinematic and engaging.
Log in or sign up for Devpost to join the conversation.