AI-Powered Campaign Creator

Table of Contents

• Objective
• Architecture Overview
  • 🔹 Core Modules
• LLM Model Selection
• Deliverables
• Getting Started & Running the Code
• Future Scope
• Contact

---

Objective

This project demonstrates a minimal viable product (MVP) for automating the creation of digital ad campaigns using a Large Language Model (LLM) and deploying them on platforms like Meta, TikTok, or LinkedIn. The tool takes a structured campaign brief and optionally a PDF input, and generates ad copy and targeting suggestions using prompt-engineered LLM responses.

---

Architecture Overview

User Input using Flask:

Campaign Details:

• Product: Smartwatch X
• Goal: Drive website traffic
• Audience: Young professionals aged 25–35 interested in fitness and productivity
• Budget: $2000
• Platform: LinkedIn
• Company Website: -
• Upload Reference PDF (Optional): -

Output of LLM

Final Ad posted on LinkedIn:

Core Modules

1. PDF Text Extraction
   Utilizes PyMuPDF for multi-page PDF parsing and conversion to usable plain text.

2. LLM-Powered Ad Generator
   Uses Google's Gemini 1.5 Flash via the Generative AI API to generate structured campaign content with tailored prompt engineering for LinkedIn.

3. Campaign Deployment (Prototype)
   Placeholder logic that simulates campaign submission. Modularized to integrate Meta, LinkedIn, and TikTok Ads APIs in production.

---

LLM Model Selection

Model	Pros	Cons	Reason for Not Choosing
Gemini 1.5 Flash	Fast, structured output, stable formatting	API limits, Easy to deploy	✅ Chosen for MVP
LLaMA 2	Open-source, local control	Requires GPU, setup-heavy	Not practical for quick MVP
Mistral	Lightweight, open license	Lower instruction-following ability	Less robust for prompt structure
Claude	Strong reasoning and summarization	Inconsistent formatting, slower	Good alternative, not ideal here
GPT-4	Powerful and accurate	Expensive, slow	Overkill for MVP, high latency

Gemini 1.5 Flash was chosen for its optimal trade-off between inference speed, formatting reliability, and ease of integration for a structured prompt environment.

---

Getting Started & Running the Code

1. Setup Environment

Clone the repository and install the required Python packages:

git clone https://github.com/your-username/atomicads-ai-engineer-assignment-margi.git
cd atomicads-ai-engineer-assignment-margi
pip install -r requirements.txt

2. Configure API Access

Create a .env file in the root directory and add your Google Generative AI API key:

GOOGLE_API_KEY=your_google_genai_api_key_here

⚠️ You need access to Google’s Gemini Generative AI API for this to work.

3. How to Set Up the Workflow in Zapier

1. Go to the Zapier Dashboard
2. Click on "Create Zap"

Step 1: Set Up Trigger

• App: Webhooks by Zapier
• Trigger Event: Catch Hook
• Zapier will generate a custom webhook URL. Use it in your Python script to POST the generated LinkedIn ad content.

4. Run the Application

Run the python app.py to generate an ad campaign:

python app.py

---

Deliverables

• Python scripts (main.py, llm_generator.py, etc.)
• Environment config (.env)
• README documentation
• Prompt examples
• Simulated ad output and screenshots

---

Future Scope

• ✅ Add frontend via Streamlit or Flask
• ✅ Real-time deployment via Ads API (Meta, LinkedIn, TikTok)
• ✅ Feedback loop and A/B testing capabilities
• ✅ Cost optimization via local open-source models
• ✅ Prompt tuning and output validation logic

---

🙋‍♀️ Contact

Feel free to reach out on LinkedIn or Email for any questions or feedback.