AI Campaign Copilot
AI-native mini CRM for building audiences and campaigns
Marketers need an analyst and SQL just to build an audience.
A CRM where a sentence becomes an audience, plus campaign drafting and simulated delivery.
From “who should we reach?” to campaign analytics in one product.
- Role
- Full-stack development
- Year
- 2026
- Stack
- Next.js, TypeScript, Express, Gemini, Zod, Tailwind CSS, Recharts
Overview
A CRM for marketers that turns a sentence like "customers who spent over ₹5,000 and haven't bought in 60 days" into an audience, drafts personalised campaigns, simulates delivery and reports performance.
The problem
Marketers know who they want to reach but building that segment needs a data analyst and SQL. Campaign tools then treat delivery and analytics as separate products.
What I built
- 1Built a monorepo with a Next.js dashboard, an Express CRM API and a separate channel service that simulates asynchronous delivery callbacks.
- 2Used Gemini to turn plain-language descriptions into audience rules and to suggest next actions, with deterministic fallbacks so the product stays demoable without a key.
- 3Seeded 1,000 customers and 5,000 orders on first boot so every screen has realistic data.
Highlights
- Natural-language audience builder.
- Delivery lifecycle simulated by an independent service, as in production messaging systems.
How it works
- Next.js dashboard
- Express CRM API
- Gemini: audience rules, suggestions
- Channel service
- Delivery callbacks → analytics
Key decisions and trade-offs
Works without an AI key
Deterministic fallbacks keep every screen demoable when Gemini is not configured.
Delivery as a separate service
An independent channel service sends asynchronous callbacks, the way real messaging providers do.
Screens
