AdMind

Multi-Agent Ad Campaign Optimizer

Built by Vivek Yadav

Overview

AdMind is a full-stack multi-agent AI application that analyzes paid advertising campaign data from a CSV file and turns it into a practical optimization report.

The app simulates how a real growth team might review Google Ads or Meta Ads performance: one agent audits the account, another creates a strategy, and a third rewrites weak ad copy. Results stream live into a clean React interface so users can watch the analysis build step by step.

This project is designed as a portfolio-ready AI engineering project: it combines agent orchestration, structured AI outputs, FastAPI streaming, CSV analysis, and a polished frontend.

What AdMind Does

AdMind accepts ad campaign CSV data and produces:

• A performance audit with spend, revenue, ROAS, wasted spend, and issue severity
• Strategic recommendations such as pausing keywords, reallocating budget, and testing copy
• Before/after ad copy rewrites for underperforming keywords
• A live visual pipeline showing each agent as it runs
• A sample-data workflow so anyone can test the product immediately

Agent Pipeline

CSV Upload
   |
   v
Agent 1: Campaign Auditor
   |
   v
Agent 2: Strategy Advisor
   |
   v
Agent 3: Ad Copywriter
   |
   v
Final Optimization Report

Agent 1 - Campaign Auditor

Finds performance problems such as:

• Wasted spend
• Low CTR keywords
• Zero-conversion keywords
• High CPC issues
• Underperforming campaigns or ad groups

Agent 2 - Strategy Advisor

Turns the audit findings into prioritized actions:

• Pause inefficient keywords
• Reduce bids on costly traffic
• Reallocate budget toward stronger campaigns
• Increase budget for high-ROAS areas
• Recommend new copy tests

Agent 3 - Ad Copywriter

Creates improved ad copy for weak keywords:

• Better headlines
• Stronger descriptions
• Clearer calls to action
• Copy aligned with search intent

Key Features

• CSV upload with validation
• One-click sample data
• Real-time Server-Sent Events streaming
• Multi-agent backend pipeline
• Structured Pydantic schemas for AI outputs
• FastAPI backend with sync and streaming endpoints
• React + Tailwind dashboard UI
• Gemini API integration with mock fallback for quota-safe demos

Tech Stack

Layer	Technology
Frontend	React 18, Vite, TailwindCSS, lucide-react
Backend	Python, FastAPI, Pydantic
AI	Google Gemini API
Streaming	Server-Sent Events
Data Input	CSV upload and sample CSV
Runtime	Stateless, in-memory pipeline

Project Structure

AdMind/
  backend/
    agents/
      auditor.py
      strategist.py
      copywriter.py
      pipeline.py
    models/
      schemas.py
    services/
      csv_parser.py
      gemini.py
    main.py
    sample_data.csv
    requirements.txt

  frontend/
    src/
      components/
      services/
      App.jsx
      main.jsx
    vite.config.js
    tailwind.config.js

  assets/
    admind-logo.svg

Getting Started

1. Backend

cd backend
.\.venv\Scripts\python.exe main.py

If you are setting up from scratch:

cd backend
pip install -r requirements.txt
copy .env.example .env
python main.py

Update backend/.env:

GEMINI_API_KEY=your_key_here
GEMINI_MODEL=gemini-2.0-flash
GEMINI_USE_MOCK_FALLBACK=true

GEMINI_USE_MOCK_FALLBACK=true keeps the demo working if Gemini quota is unavailable.

2. Frontend

cd frontend
npm.cmd install
npm.cmd run dev

Open:

http://localhost:5173

Then click Try Sample Data.

API Endpoints

Method	Endpoint	Purpose
GET	/health	Backend health check
GET	/sample-csv	Download sample campaign CSV
POST	/analyze	Run pipeline with SSE streaming
POST	/analyze-sync	Run pipeline and return one JSON response

Why This Project Matters

AdMind demonstrates how agentic AI can be applied to a practical business workflow. Instead of asking one generic model to do everything, the system separates the work into specialized agents. This makes the pipeline easier to reason about, easier to test, and closer to how production AI systems are designed.

It also shows the complete product loop: upload data, process it through AI agents, stream progress live, and present results in a polished interface.

Current Status

Completed:

• Backend foundation
• CSV parser
• Gemini service
• Three-agent pipeline
• FastAPI routes
• SSE streaming
• React frontend
• Sample data workflow
• Professional README and branding

Author

Vivek Yadav

Project: AdMind - Multi-Agent Ad Campaign Optimizer