TubeGuardAI 🛡️

TubeGuardAI is an producion-grade Automated Video Compliance QA Pipeline. Orchestrated by LangGraph, it is designed to audit video content against complex regulatory standards using a Multimodal RAG (Retrieval-Augmented Generation) architecture.

By synthesizing visual cues, transcripts, and OCR (Optical Character Recognition) data from video, TubeGuardAI transforms unstructured video into structured, actionable compliance reports with production-grade observability.

This project uses tools from Microsoft Azure: Cloud Computing Services

---

🚀 Key Features

• Multimodal Ingestion: Extracts deep insights (transcripts, OCR, and visual Cues) from video using Azure AI Video Indexer.
• Agentic Orchestration: Leverages LangGraph to manage complex, stateful reasoning flow for auditing.
• Intelligent Retrieval: Uses Azure AI Search and Azure OpenAI (Embedding model) to pull relevant regulatory rules based on video context.
• Deterministic Reasoning: Powered by Azure OpenAI (LLM) to detect violations and generate precise, evidence-based reports.
• Full-Stack Observability:
  • LangSmith: Granular tracing for LLM workflow optimization.
  • Azure Application Insights: Real-time telemetry, logging, and performance monitoring.

API Layer (FastAPI)

TubeGuardAI exposes its agentic workflow through a high-performance FastAPI backend, enabling seamless integration with frontend applications and enterprise dashboards.

Key Features:

• Asynchronous Orchestration: Designed to handle long-running video AI tasks using non-blocking execution patterns.
• Pydantic Data Validation: Strict schema enforcement for AuditRequest and AuditResponse models, ensuring 100% Type Safety.
• Production Telemetry: Integrated with Azure Application Insights via a custom telemetry module to track request latency, success rates, and system health.
• Automatic Documentation: Interactive Swagger UI and ReDoc generated automatically for rapid developer onboarding.
• Robust Error Handling: Centralized exception management using FastAPI HTTPException to provide actionable feedback to clients.

API Endpoints:

Method	Endpoint	Description
POST	/audit	Triggers the LangGraph compliance workflow for a given video URL.
GET	/health	Service heartbeat and health check.

---

🏗️ System Architecture

The pipeline follows a sophisticated workflow to ensure high-fidelity compliance checks:

1. Data Ingestion(Azure Blob Storage): Raw video files are uploaded to an Azure Blob Storage container.
2. Data Extraction: Videos are processed through Azure Video Indexer to generate Transcription, OCR data and Visual Cues.
3. Regulatory RAG: The system queries a vector database of compliance documents to find rules applicable to the specific video content.
4. Audit Logic (LangGraph): An agentic flow evaluates the video data against the retrieved rules from pdfs files.
5. Structured Output: The final audit is delivered as a JSON Compliance Report, ready for downstream integration.

---

🛠️ Tech Stack

Component	Technology
Orchestration	LangGraph
Storage	Azure Blob Storage
LLM	Azure OpenAI ('GPT-4o')
Video Intelligence	Azure AI Video Indexer
Vector Database	Azure AI Search
Embeddings	Azure OpenAI ('text-embedding-3-small')
Observability	LangSmith & Azure Application Insights
Language	Python 3.12

---

📋 Prerequisites

Before running the pipeline, ensure you have the following:

• An active Azure Subscription.
• Azure OpenAI deployment (GPT-4o and Embedding model).
• Azure AI Search service instance.
• Azure Video Indexer account.
• LangSmith API Key for debugging and tracing.

---

⚙️ Installation & Setup

This project uses uv as the python package manager

Install uv:

curl -LsSf https://astral.sh/uv/install.sh | sh

Verify the installation:

echo 'export PATH="$HOME/snap/code/221/.local/bin:$PATH"' >> ~/.bashrc && source ~/.bashrc

uv --version

Clone the repository or Download as zip file

https://github.com/yoursrealkiran/TubeGuardAI.git

cd TubeGuardAI

Environment Setup

Create a Virual Environment

In the terminal, run the below command to create a virtual environment

uv venv

Activate the environment

source .venv/bin/activate

Install Dependencies

uv sync

Running the Application

1. To run the script

Run the script to begin the audit:

uv run python main.py

2. To run using FastAPI backend and to see output in Swagger UI

uv run uvicorn backend.src.api.server:app --reload

Open /audit endpoint in Swagger UI, enter YouTube URL and execute to begin audit.