Catalyst

Inspiration

Marketers today face a fragmented workflow — jumping between trend research, content generation tools, design software, translation apps, and scheduling platforms. We wanted to streamline this process with an AI-first, end-to-end campaign assistant that turns goals into campaigns in minutes, not weeks.

What it does

Catalyst is a multimodal AI-powered marketing assistant that helps SMBs go from idea to execution. Users input campaign goals or a creative brief. Catalyst then:

Gathers real-time market trends via Apify
Analyzes competitor content
Suggests campaign strategies and ICPs
Generates draft copy and visuals (with Vizcom)
Translates copy via DeepL
Schedules posts and outreach via Arcade and Vapi

All within a unified dashboard, with full user control and editing.

How we built it

Frontend: React + TypeScript using ShadCN for a responsive dashboard UI
Backend: Python FastAPI with async APIs for agent orchestration
Agents: Each functionality is handled by modular agents (Market Research, Content Gen, Scheduler, etc.) communicating via a central Multi-Agent Control Plane (MCP)

Integrations:

Apify for scraping trends and competitor ads
Vizcom for sketch-to-render ad visual generation
DeepL API for translation
Arcade for publishing/scheduling content
Vapi for outreach call handling

Data Sources:
Google Trends, Facebook Ad Library, Unsplash API (optional)

UX:
Whiteboard-style editable UI, campaign calendar, and real-time previews

Challenges we ran into

Coordinating async responses from multiple agents without breaking the user flow
Handling visual input (e.g., sketches) and converting them into polished visuals reliably
Balancing automation with human editability — ensuring all outputs remained editable without breaking sync
Integration quirks: managing API limits and varied response formats from sponsor tools

Accomplishments that we're proud of

Successfully orchestrated a multi-agent backend with real-time communication
Generated full campaign outputs — text, visuals, translations, and scheduling — from a single input prompt
Built a functional whiteboard UX with drag-and-drop refinement of content
Demonstrated real sketch-to-render ad visual flow using Vizcom

What we learned

Modularizing AI workflows as autonomous agents vastly improves maintainability and clarity
Multimodal input/output adds huge UX value but requires careful interface design
The value of integrating industry tools like Apify and DeepL directly into an AI workflow
How marketers balance creative control with automation — and how to support both

What's next for Catalyst

Add support for live A/B testing and performance feedback (via Arize AI)
Integrate email and CRM systems for lead capture and nurture
Build more intelligent ICP/segment discovery using historical data and clustering
Expand to handle video creatives and TikTok-style shortform generation
Fine-tune LLM prompts per industry or brand tone to improve brand voice control
Launch a beta with real SMBs to test real-world marketing workflows