AI Department – Project Story
🌟 Inspiration
Clients kept asking for a “fractional CMO” or some other AI powered executive team member. I realized I could simulate an entire growth department with specialized AI agents that collaborate on demand.
📚 What I Learned
- Role-specific prompts outperform one generalist LLM.
- Users love PDF hand-offs for board decks.
- Chunking uploads (< 2 k tokens) prevents context dilution during retrieval.
🛠️ How I Built It
| Layer | Tech & Flow |
|---|---|
| Front-end | React + Vite dashboard, Tailwind UI. |
| Auth & Storage | Supabase Auth; file uploads stored & indexed for RAG. |
| Agent Orchestration | LangChain routes the prompt through five OpenAI o4-mini agents (researcher, strategist, copywriter, ops, marketer). |
| Retrieval | Embeddings created on upload; each agent can call get_context() to cite docs. |
| Export | React-PDF + Serverless Puppeteer compile chat into branded PDF. |
| Hosting | Netlify CI/CD with edge functions for agent pipeline. |
⚠️ Challenges
| Issue | Fix |
|---|---|
| Long agent response times | Parallel async calls with streaming back to UI. |
| Prompt contamination between roles | Isolated memory chain per agent + system guardrails. |
| File upload size vs. Netlify limits | Chunked uploads and background processing queue. |
🚀 What’s Next
Slack plugin – run agents directly in team channels.
Agent marketplace – community-built specialties (e.g., SEO wizard, Data analyst).
AI Department is on track to become the plug-and-play growth engine that scales with every stage of a business.
Built With
- netlify
- openai
- puppeteer
- react
- supabase
- tailwind
- vite
Log in or sign up for Devpost to join the conversation.