Posts

How Generative AI works?

Image
This simple picture illustrates the fundamental architecture of how Generative AI systems work. It shows a three-stage process: 1) Input Stage (Data): The foundation model is trained on diverse data types including text, images, speech, structured data, and 3D signals. This comprehensive training data gives the model broad knowledge and capabilities. 2) Core Processing: The "Foundation Model" sits at the center, representing the large-scale AI model that has learned patterns and relationships from all the training data. 3) Output Stage (Tasks): Through a process called "Adaptation," the foundation model can be fine-tuned or prompted to perform specific tasks like question answering, sentiment analysis, information extraction, image captioning, object recognition, and instruction following. The key insight this diagram conveys is that modern GenAI systems use a single, versatile foundation model that can be adapted for multiple different applications, rather than ...

4 weeks of my GenAI Capstone project plan

Image
As part of PGP AIFL (Post Graduate Program in Artificial Intelligence for Leaders) from University of Texas Austin, we need to submit a Capstone project with other 4 members assigned by program coordinator. This project has 3 parts and each part has a weightage of marks given below. The pass mark for this project is 60% to finish the 5 mons of this program started in Jan'2025. Part-1:  High level idea of what we are designing & implementing the project (25 marks) Part-2: Detailed document on the project implementation (35 marks) Part-3: Final presentation deck with the group (40 marks) Initially we started with 3 ideas in our mind to discuss & finalize as a group.  Car search using AI/ML Real estate Stock market  HR Assist Intake Accelerator Application (IAA) - Telecom domain During our kickoff meeting, we all agreed to take up an existing live HR assistant application using GenAI platform & RAG model. We started focusing on the requirements, data, design, s...

What a Transformer in AI is?

Image
Transformers in AI are a type of artificial neural network architecture that is used to solve the problem of transformation of input sequences into output sequences in deep learning applications.  Transformers consist of an encoder-decoder architecture, where the encoder processes the input sequence and the decoder generates the output sequence. Each encoder and decoder layer in the transformer contains multiple self-attention mechanisms followed by feedforward neural networks.  The following is a very simple representation of a transformer. The two components in the architecture: the encoder and decoder. The former is the element that receives each component of the input sequence, encoding (hence the name) it into a vector carrying the context information about the whole sequence. This is sent to the decoder whose job is to understand the context and resolve the output meaningfully. Here is a simple translation of English sentence to German. Behind the scenes, the job of the ...

MCP integration allows Claude to store the content locally

Image
  Brief idea on MCP: MCP( 𝗠𝗼𝗱𝗲𝗹 π—–π—Όπ—»π˜π—²π˜…π˜ π—£π—Ώπ—Όπ˜π—Όπ—°π—Όπ—Ή) is an open standard developed by Anthropic to facilitate seamless integration between AI assistants and external data sources, tools, and environments. Traditionally, connecting AI systems to various data sources required custom implementations for each integration, leading to scalability issues. MCP offers a universal protocol that simplifies these connections, allowing AI systems to access the data they need more efficiently. π—žπ—²π˜† π—–π—Όπ—Ίπ—½π—Όπ—»π—²π—»π˜π˜€ 𝗼𝗳 𝗠𝗖𝗣: 1. 𝗠𝗖𝗣 π—¦π—²π—Ώπ˜ƒπ—²π—Ώπ˜€ : These expose data and functionalities from various sources, such as content repositories, business tools, and development environments, making them accessible to AI applications. 2. 𝗠𝗖𝗣 π—–π—Ήπ—Άπ—²π—»π˜π˜€: AI applications that connect to MCP servers to retrieve data or perform actions, enabling seamless interaction with external systems. Anthropic has open-sourced MCP, providing developers with specifications, SDKs, an...