Karthick's Sunday Learning (16/11)

As I strive to learn every day, here are my this week's learning and deep reading as I took some time to recap/reflect on those topics for my network and newsletter. Are you motivated to learn more every day? If not, ask yourself what is missing and go after it.

WebRTC

While preparing for the Azure AI-102 exam, I came across WebRTC in the context of real-time communication scenarios. The exam emphasises building intelligent solutions on Azure, and one key area is enabling live audio/video interactions for applications like virtual assistants, telehealth, and customer support.

This led me to explore how Azure Communication Services integrates WebRTC for browser-based calls, signaling, and NAT traversal using TURN/STUN servers. I realised that WebRTC isn’t just about video calls, it’s a foundational technology for low-latency, secure, and scalable real-time experiences, which are increasingly relevant in AI-driven solutions.

WebRTC (Web Real-time communication) is an open-source technology and API standard that enables direct peer-to-peer communication between browsers and mobile applications. It allows for real-time audio, video, and data sharing without the need for plugins or external software.

• It’s like FaceTime or WhatsApp video, right in your browser.
• You open a web page, and you can instantly talk, see, or send stuff to someone else.
• Everything is secure and private, and it works super fast because it connects people directly.

Under the hood

Let us take a look under the hood of WebRTC. Following are the main technical details:

Core Components

• MediaStream API: Captures audio and video from microphones, cameras, or screens.
• RTCPeerConnection API: Establishes and manages the peer-to-peer connection, handling the exchange of media and data between clients.
• RTCDataChannel API: Allows direct transfer of arbitrary data (e.g., files, chat messages) between peers.

How Connections Work

• Signaling: WebRTC does not define a signaling protocol; applications must use their own (commonly WebSockets or HTTP) to exchange connection setup messages, offer, answer, and ICE candidates between peers.
• Session Description Protocol (SDP): Used to describe media capabilities (e.g., codecs, resolutions) and connection information during setup.
• ICE (Interactive Connectivity Establishment): Protocol for finding the best path between peers, including NAT traversal.

Protocols and Standards

• DTLS (Datagram Transport Layer Security): Encrypts control messages and all media/data streams for privacy and security.
• SRTP (Secure Real-Time Transport Protocol): Encrypts and transports audio/video streams.
• RTP/RTCP (Real-Time Transport Protocol / Real-Time Control Protocol): Used for packet delivery and QoS reporting.

Simple flow

1. Signaling → Exchange “hello” info
2. ICE → Find best route
3. DTLS → Secure the connection
4. SRTP → Send audio/video
5. SCTP → Send data/messages

Key features

• Real-time communication for voice, video, and file/data transfer.
• Peer-to-peer architecture means connections are direct between users, improving speed and privacy.
• Built into most modern browsers (Chrome, Firefox, Edge, Safari) and supported by various platforms.
• Uses encryption and NAT traversal techniques (STUN, TURN, ICE) to maintain connection security and adapt to complex networks.
• Common use cases: video chat (like Google Meet or WhatsApp Web calls), screen sharing, collaborative tools, online gaming, and secure file transfer.

Why it matters

• Why it’s important:

* Instant Connectivity: Works directly in browsers and mobile apps.

* Low Latency: Perfect for live video, gaming, and remote collaboration.

* Secure: Built-in encryption for audio, video, and data.

* Open Standard: Free and widely supported by major browsers.

• Everyday Examples:

* Video meetings (like Teams, Zoom)

* Telehealth consultations

* Customer support with live video

* IoT devices streaming sensor data

In short, WebRTC makes real-time communication fast, secure, and easy, right in your browser or app

WebRTC in Azure

WebRTC on Azure enables secure, real-time audio, video, and data communication directly in browsers and apps. Azure supports this through Azure Communication Services (ACS), which provides built-in signaling, TURN/STUN servers for NAT traversal, and a scalable SFU for multiparty calls.

• For custom solutions, developers can deploy their own signaling servers, media relays, and analytics pipelines using Azure App Service, Functions, Kubernetes (AKS), and Blob Storage. This approach ensures low latency, global reach, and enterprise-grade security with Azure Active Directory integration.

Hugging Face

If you’ve been following AI trends, you’ve probably heard of Hugging Face. It’s more than just a library. It is a global hub for state-of-the-art machine learning models, datasets, and tools.

Hugging Face is a leading AI company and open-source platform best known for its work in natural language processing (NLP), machine learning, and AI model sharing.

• It’s like a “library” where anyone can find and use smart computer programs that work with words, pictures, and language.
• You can search for a model (like one that translates languages or writes stories), try it out, and share your own creations.
• Lots of people, from students to big companies, use it to make apps and tools smarter, faster, and more helpful.

If you want a robot to answer questions, write text, or recognise faces in photos, Hugging Face makes it much easier to find the right model to your started, you don’t need to build everything from scratch!

It is consists of:

• Model Hub: A huge repository of pre-trained AI models for tasks like text analysis, translation, image recognition, and speech.
• Transformers Library: The most popular Python library for working with advanced models like BERT, GPT, and T5.
• Community: A global space where developers and researchers share models, datasets, and tools.
• Deployment Tools: APIs and integrations with cloud platforms (Azure, AWS, GCP) to easily use and deploy AI models.

Key points about HF

• Model Hub: Hugging Face hosts a massive online repository (“Model Hub”) of thousands of pre-trained AI models for tasks like text generation, translation, summarisation, question answering, image processing, and more. These models include BERT, GPT, LLaMA, and specialised community models.
• Transformers Library: Their widely-used open-source library, transformers, allows developers to easily use and fine-tune state-of-the-art models from companies like Google, Meta, OpenAI, and others in their own applications, using Python and standard deep learning frameworks (PyTorch, TensorFlow).
• Community & Collaboration: Hugging Face makes it easy for researchers and developers to share, discover, and collaborate on models, datasets, and AI workflows - accelerating innovation and reproducibility.
• Spaces & Inference Endpoints: You can deploy interactive machine learning apps directly on Hugging Face (called “Spaces”) and turn models into production APIs without having to manage your own infrastructure.
• Other Tools: The platform offers tools for data labelling, evaluation, secure model deployment, and tracking.

Why it matters

Hugging Face has democratized AI development, enabling fast, collaborative, and open innovation in NLP and broader machine learning. It’s a core resource for anyone building with AI today, from hobbyists to large enterprises.

• Democratizes AI: Anyone can access state-of-the-art models without needing massive resources.
• Speeds up development: Pre-trained models and easy-to-use libraries save months of work.
• Community collaboration: Researchers and developers share models, datasets, and best practices globally.
• Supports multiple domains: NLP, computer vision, speech, and multimodal AI - all in one ecosystem.
• Enterprise ready: Integrates with cloud platforms like Azure, AWS, and GCP for production deployment.

How to get started

To get started with Hugging Face, follow these simple steps:

1. Visit the Website Go to huggingface.co - this is where you’ll find all the models, datasets, and tools.

2. Explore or Search Look for a model you are interested in, like one for text generation, translation, or image recognition. You can try many models directly in your browser, no coding needed.

3. Create a Free Account Sign up with your email if you want to save favourites, upload your own models, or use more advanced features.

4. Try Models Online Click on a model, type in your own text or upload a file, and see how it works. Many models have interactive demos.

5. Use in Your Projects (optional) If you know Python (a common coding language), you can install the Hugging Face library on your computer:

<run in bash>
pip install transformers        

Then, follow tutorials on the Hugging Face site to use models in your own scripts or apps.

6. Join the Community Visit the forums, read docs, and check out guides for learning tips and project ideas. The Hugging Face community is very supportive and beginner-friendly!

Real-world example

Here’s a real-world example of using Hugging Face:

Suppose you work for a company that receives thousands of customer emails every day. Instead of manually reading and replying to each one, you want to automatically sort emails and generate helpful answers using AI.

How Hugging Face helps

• Automatic Email Sorting (Classification)
• Use a Hugging Face model to scan incoming messages and instantly tag them as “support,” “sales,” “complaint,” or “question.”
• This saves time and routes each email to the right department.
• AI-Powered Reply Generator (Text Generation)
• Connect your email system to a Hugging Face model (such as GPT-3, GPT-4, or Llama-2).
• When a customer asks something, the model drafts a polite, accurate reply automatically.
• Example: If a customer asks, “How do I reset my password?” The AI can respond with: “To reset your password, click ‘Forgot Password’ on our login page and follow the instructions. If you need help, reply to this email.”
• Language Translation
• If you receive emails in different languages, Hugging Face offers models that can instantly translate messages so your team can understand and reply in English (or any target language).
• Sentiment Analysis
• Find out which customers are unhappy or need urgent help, based on the tone of their messages.

How to set it up

• Connect your company’s email system to these Hugging Face models (using simple code or integrations).
• The system runs automatically, sorting, analysing, translating, and replying to emails, all powered by AI.

As an end result, your team responds faster, gets less overwhelmed, and customers receive timely, accurate support. This is just one way Hugging Face is used in the real world for businesses, saving time and improving service!

This is it for this week. I will be back next Sunday with a few more interesting topics for us to learn together.

What are you learning? May be we can learn together!

Happy learning - Keep learning and stay hungry!