CampaignCrafter - AI-Powered Ad Campaign Generator

Architecture diagram

Inspiration

Creating a compelling ad campaign usually demands hours of manual effort—researching trends, drafting copy, designing creatives in tools like Canva, and localizing content for diverse audiences. This scattered and repetitive process becomes a bottleneck, especially for startups or solo marketers who need to launch quickly and adapt fast. We were inspired to remove this friction and build a system that automates the entire ad creation process end-to-end using AI agents.

What it does

CampaignCrafter is an intelligent, multi-agent system that automates the creation of complete ad campaigns from a simple prompt. It:

Collects market insights from Google and Instagram
Generates platform-specific ad copy (e.g., for Instagram, Google Ads, etc.)
Creates visuals using AI image generation and stores them in Google Cloud Storage
Translates the copy into a user-selected language
Allows users to edit and save the final campaign to their dashboard

The entire campaign is generated in under a minute, enabling marketers to go from idea to execution instantly.

How we built it

CampaignCrafter is built using Google’s Agent Developer Kit (ADK). Using ADK, we defined a parent agent that orchestrates four specialized child agents:

Market Insight Agent – Scrapes web + Instagram data to extract hashtags, trends, and audience insights.
Copywriter Agent – Writes SEO-friendly ad copy using the Gemini model.
Image Generation Agent – Creates engaging ad visuals using Gemini model.
Localization Agent – Translates the final copy into the target language using Google Cloud Translate.

Tech Stack: Frontend: Next.js, TailwindCSS, Firebase Auth

Backend: Python (Flask), ADK

Storage: Google Cloud Storage, Firestore

AI Models: Gemini 2.0 Flash , text-to-image model (Gemini 2.0 Flash Exp)

APIs: SERP API, Google Translate

The UI supports:

Prompt input with platform specification
Language selection
Campaign preview
Editing and saving campaigns

Challenges we ran into

Prompt Engineering: Tuning prompts for Gemini to deliver platform-specific, brand-consistent copy took significant iteration.
Image Generation: Achieving ad-quality visuals, generating descriptive prompt for image generation by agent itself.
Trends Scraping: Search engine scraping needed to be lightweight and reliable, with caching to avoid blocks.
Managing memory and timeout issues when multiple agents were running in sequence
Agent Orchestration: Maintaining smooth data flow and state across multiple agents required careful design, but ADK’s session state features helped streamline this.

Accomplishments that we're proud of

Built a fully automated, multi-agent ad campaign generator from scratch that mirrors the actual flow of a creative marketing team
Achieved seamless orchestration between agents using Google’s ADK
Delivered a user-friendly frontend that produces campaign assets in seconds
Enabled multilingual, platform-optimized ad creation with just one prompt

What we learned

How to build and orchestrate multi-agent systems using ADK
Importance of controlling agent instruction precision to avoid tool misuse.
Efficient data flow and session management across async agents - ADK’s session state features were invaluable
Handling production deployment and memory optimization for agent tools