🌩️ AI-driven Cloud Cost Copilot (FinOps Assistant)

---

📌 Overview

The FinOps Assistant is an AI-driven cloud cost analytics platform designed to help organizations monitor, analyze, and optimize their cloud spending. It provides:

• End-to-end pipeline: ETL → KPIs → RAG → Recommendations → API → UI
• KPI dashboards to analyze cost trends
• Natural-language Q&A over your cost data and FinOps documentation
• Actionable recommendations to reduce costs

---

🧩 Features

• Cost Analysis — Monthly spend trends, top cost drivers, cost anomalies
• RAG Q&A — Ask natural questions like “Why did compute cost increase in May?”
• Recommendations — Detect idle resources, tagging gaps, sudden spikes
• UI + API — Interactive Streamlit dashboard + FastAPI backend
• Evaluation Suite — Measure retrieval quality (Recall@k) + answer quality

---

🧠 Architecture

---

🗂️ Directory Structure

├── app/
│   ├── main.py             # FastAPI backend entrypoint
│   ├── models.py            # SQLAlchemy schema (billing, resources)
│   ├── analytics.py         # KPI & trend calculations
│   ├── rag.py               # RAG pipeline setup (retriever + LLM)
│   ├── etl.py               # ETL piepline to validate data
|   ├── reccomendations.py   # Suggestions Generation
|   ├── rag_qa.py            # LangChain QA chain logic
│   └── validators.py        # input validation, prompt-injection guard 
|                        
├── docs/                     # Project documentation
│   ├── finops.md             # Refernce Docs for LLM
│   ├── PRD.pdf               # Product requirement Document
│   ├── Technical Design doc.pdf
│   ├── Assets/              
|
├── UI/
│   └── app.py                # Streamlit frontend dashboard
├── scripts/
│   ├── ingest_billing.py
│   ├── seed _resources.py
│   ├── build_faiss_index.py
│   ├── update_assignments.py
|   ├── manage_data.py
│   └── ...
|
├── tests/                    # Testing Files for evaluation
│   ├── conftest.py
│   ├── evaluate_rag.py
│   ├── test_analytics.py
│   ├── test_reccomendations.py
|   ├── test_reccomendations_api.py
│   └── rag_eval_results.json
|
|
├── data/                     # SQLite DB (billing.db) & Data Files
├── requirements.txt
├── Dockerfile
├── docker-compose.yml
├── .dockerignore
├── .env.example
└── README.md

---

⚙️ Tech Stack

Layer	Technology	Purpose
ETL & DB	Python, SQLAlchemy, SQLite	Data ingestion + schema
Analytics	Pandas, SQL	KPI & cost trend calculations
RAG Pipeline	sentence-transformers, FAISS, LangChain, Groq LLM	Natural-language Q&A
API Layer	FastAPI	Serve KPIs and Q&A endpoints
Frontend	Streamlit	Interactive dashboard UI
Deployment	Docker, Docker Compose	Containerization
Observability	LangSmith	Logging + traces
Security	Prompt-injection guards, keyword filters	Secure LLM input

---

🧪 Setup & Running Locally

1. Clone repo

   git clone https://github.com/jasoncobra3/FinOps-Copilot.git
   cd FinOps-Copilot
   

2. Create environment

   python -m venv .venv
   source venv/bin/activate  # or venv\\Scripts\\activate (Windows)
   

3. Install dependencies:

   pip install -r requirements.tx
   

4. Initialize DB

     python -m app.models
   

5. Generate Sammple Data

   python scripts/generate_sample_data.py
   

6. Ingest sample data

   python scripts/ingest.py  --input data/sample_billing.csv
   python scripts/seed_resources.py
   python scripts/update_assignments.py ##generate randomness in data
   

7. Build RAG index

   python scripts/build_faiss_index.py
   

8. Run FAstAPI Backend

   uvicorn app.main:app --reload
   

• FastAPI - http://localhost:8000

12. Run Streamlit Frontend

    streamlit run UI/app.py
     

• Streamlit - http://localhost:8501

---

📊 Evaluation

Run RAG retrieval evaluation

   python tests/eval_rag.py

• Outputs Recall@1/3/5 scores in CSV
• Also includes answer quality scores (1–5 rubric)

---

🧠 Design Decisions & Trade-offs

• SQLite chosen for local simplicity, can migrate to PostgreSQL later
• FAISS is fast and lightweight for demo scale (vs hosted vector DBs)
• LangChain simplifies orchestration but adds abstraction → documented clearly
• Streamlit is quick to build for MVP, though not production-grade
• Groq LLM chosen to avoid dependency on paid APIs

---

🚀 Future Work

• Migrate to PostgreSQL for multi-user deployments
• Add real billing data ingestion from cloud APIs
• Implement cost optimization engine (idle resource detection, right-sizing)
• Add user auth + role-based access control
• Enhance Streamlit UI (filters, charts, reports)
• Deploy on Railway or Render with CI/CD

---

📸 Screenshots

Dashboard KPI View	Month DropDown	Reccommendations

Underutilized resources	AI RAG Chatbot	Response With Resource group

---

🌟 Contributing

Feel free to fork, star, or submit a pull request to contribute improvements!