AdOps AI Copilot

AI-powered AdTech diagnostics platform that uses Retrieval-Augmented Generation (RAG), vector search, and Gemini LLM to analyze campaign performance, detect issues, and generate optimization recommendations.

---

Live Demo

Frontend: VERCEL_LINK
Backend API Docs: HUGGINGFACE_BACKEND_URL/docs

---

Product Preview

Dashboard

Analytics + Executive Summary

AI Copilot

PDF Report Export

---

Overview

AdOps AI Copilot is a full-stack AI product designed for marketing, growth, and Ad Operations teams.

It helps diagnose underperforming ad campaigns using real campaign KPIs, retrieves relevant historical issues and support-ticket evidence, and uses an LLM to generate grounded troubleshooting recommendations.

This project simulates how enterprise AdTech teams can reduce manual investigation time and make faster optimization decisions.

---

Features

Campaign Diagnostics Engine

• Campaign health scoring
• CTR / CPC / Conversion Rate / ROAS analysis
• Spend monitoring
• At-risk campaign detection
• KPI performance dashboard

Automated Issue Detection

• Low CTR detection
• Poor ROAS alerts
• High CPC warnings
• Weak conversion rate identification
• Budget waste spotting

Real Retrieval-Augmented Generation (RAG)

• Semantic embeddings using Sentence Transformers
• FAISS vector similarity search
• Relevant campaign context retrieval
• Ticket evidence retrieval
• Grounded LLM responses

LLM-Powered AI Troubleshooting Assistant

• Gemini-powered recommendations
• Root cause analysis
• Executive summaries
• Evidence-backed responses
• Actionable next steps

Interactive Dashboard

• SaaS-style UI
• Charts and analytics cards
• Real-time API integration
• Clean responsive design

Executive Reporting

• Downloadable PDF reports
• Stakeholder summaries
• Campaign performance snapshots

---

System Architecture

React + Vite Frontend
        ↓
FastAPI Backend API
        ↓
Campaign KPI Dataset + Support Tickets
        ↓
Sentence Transformers Embeddings
        ↓
FAISS Vector Search
        ↓
Top Relevant Context Retrieval
        ↓
Gemini LLM
        ↓
Generated Troubleshooting Answer

---

Tech Stack

Frontend

• React.js
• Vite
• Tailwind CSS
• Framer Motion
• Recharts
• Axios
• jsPDF
• html2canvas

Backend

• FastAPI
• Python
• Uvicorn
• Pydantic

AI / ML

• Sentence Transformers
• FAISS
• Gemini API
• Retrieval-Augmented Generation (RAG)

Data Processing

• Pandas
• NumPy
• scikit-learn

Deployment

• Vercel
• Hugging Face Spaces

---

Project Structure

Adops-ai-copilot/
│
├── frontend/
│   ├── src/
│   │   ├── components/
│   │   ├── api/
│   │   ├── App.jsx
│   │   └── main.jsx
│   └── package.json
│
├── backend/
│   ├── app/
│   │   ├── main.py
│   │   ├── routes/
│   │   │   ├── campaigns.py
│   │   │   ├── diagnostics.py
│   │   │   └── copilot.py
│   │   ├── services/
│   │   │   ├── analyzer.py
│   │   │   ├── rag_service.py
│   │   │   └── llm_service.py
│   │   └── data/
│   │       ├── campaigns.csv
│   │       └── tickets.csv
│   ├── requirements.txt
│   └── Dockerfile
│
├── assets/
│   ├── dashboard.png
│   ├── charts.png
│   ├── copilot.png
│   └── report.png
│
└── README.md

---

How It Works

1. Campaign performance data is loaded from structured datasets.

2. Diagnostics engine computes:

• CTR
• CPC
• Conversion Rate
• ROAS
• Health Score

3. Support-ticket and campaign records are embedded into vectors.

4. FAISS retrieves the most relevant campaign issues.

5. Gemini LLM receives grounded context.

6. AI Copilot generates recommendations.

7. User interacts through a professional dashboard.

---

Example Query

User asks:

Why is CAMP005 performing badly and what should we do?

AI Response:

Executive Summary:
CAMP005 is underperforming due to poor ROAS, low CTR, and audience fatigue.

Evidence Found:
Retrieved diagnostics and support-ticket records related to CAMP005.

Root Cause:
Weak engagement, inefficient spend, and remarketing fatigue.

Recommended Actions:
- Rotate creatives
- Expand audience pool
- Exclude recent converters
- Pause weak ad sets
- Improve landing page offer

---

Running the Project Locally

1️ Clone Repository

git clone https://github.com/Siyonova/Adops-ai-copilot.git
cd Adops-ai-copilot

2️ Frontend Setup

cd frontend
npm install
npm run dev

Frontend runs at:

http://localhost:5173

3️ Backend Setup

cd backend
pip install -r requirements.txt
uvicorn app.main:app --reload

Backend runs at:

http://127.0.0.1:8000

API docs:

http://127.0.0.1:8000/docs

---

Dashboard Features

• KPI summary cards
• Spend insights
• ROAS charts
• Campaign health graph
• Diagnostics table
• AI Copilot query panel
• PDF report download

---

Example Use Cases

• Diagnose low CTR campaigns
• Detect poor ROAS accounts
• Investigate budget waste
• Analyze conversion drop-offs
• Support AdOps teams with AI
• Generate stakeholder reports

---

Business Impact

This platform demonstrates how growth teams can:

• Reduce manual debugging time
• Detect revenue leaks quickly
• Prioritize weak campaigns faster
• Combine support feedback with metrics
• Use AI for optimization workflows

---

Future Improvements

• Google Ads API integration
• Meta Ads API integration
• Live campaign ingestion
• Multi-account dashboards
• Forecasting models
• Anomaly detection
• User authentication
• Team collaboration notes
• LLM fine-tuning on AdTech workflows

---

Author

Siyonova

Passionate about AI systems, full-stack products, analytics platforms, and enterprise ML solutions.

GitHub: https://github.com/Siyonova

---

Support

If you like this project, star the repo and connect with me.