Blogs | Tuhin Sharma

RAPTOR - A Smarter Way to Retrieve and Use Information in AI

Mon, 10 Mar 2025 14:15:00 +0000

Description

Large language models (LLMs) like ChatGPT and GPT-4 are incredibly powerful, but they struggle to keep up with new information and understand long, complex documents. Traditional retrieval methods pull short chunks of text, often missing important context. This is where RAPTOR (Recursive Abstractive Processing for Tree-Organized Retrieval) comes in — a new way to retrieve and summarize information efficiently. In this article, we’ll break down what RAPTOR is, why it’s better than traditional methods, and how to implement it with a hands-on example.

Read more..

Building a Biomedical Question-Answering System Using RAG

Mon, 03 Mar 2025 14:15:00 +0000

Description

With the explosion of biomedical research and literature, extracting relevant information efficiently has become a challenging task. Large Language Models (LLMs) like GPT-4o-mini, combined with a Retrieval-Augmented Generation (RAG) pipeline, offer a promising approach for building robust question-answering (QA) systems.

Read more..

LlamaIndex vs LangChain vs Haystack vs Llama-Stack - A Comparative Analysis

Thu, 06 Feb 2025 14:15:00 +0000

Description

As organizations increasingly integrate AI-driven search and retrieval systems into their workflows, the choice of the right framework becomes critical. Among the leading solutions, LlamaIndex, LangChain, Haystack, and Llama-Stack stand out as powerful tools for building retrieval-augmented generation (RAG) pipelines, enabling seamless interactions between Large Language Models (LLMs) and enterprise data.

Read more..

Decoding the Power of Multi-Head Attention in Transformers

Wed, 06 Nov 2024 14:15:00 +0000

Description

In our pursuit of creating a fun dialogue completer using transformers, we chose to build a system that could complete famous quotes — like those from Cersei Lannister when she played the game of thrones. To achieve this, we first passed our input text through an embedding layer, adding position information to each word embedding to create position-aware embeddings. These were then sent forward to the next layer. In this segment, we dive deeper into the transformative heart of transformer models: the multi-head attention mechanism, which, I must warn, is a powerful multi-faceted component.

Read more..

Revolutionizing Smart Buildings with Federated Learning

Sun, 19 May 2024 14:15:00 +0000

Description

In today’s world, IoT (Internet of Things) devices are integral to our daily lives. From wearables and self-driving cars to smart buildings and cities, these devices shape and control our environments. Among these innovations, smart buildings stand out by not only ensuring the comfort and safety of occupants but also promoting energy and financial efficiency. The integration of AI (Artificial Intelligence) in smart buildings has made them even smarter by leveraging data from various sensors to optimize building functionalities.

Read more..

Logo detection using Apache MXNet

Thu, 01 Feb 2018 14:15:00 +0000

Description

Given an image, identify which company’s logo is present.

The article will motivate the business problem, how to get data, how to use MXNet to build the deep learning model

Read more..

Uncovering hidden patterns through machine learning

Thu, 21 Dec 2017 14:15:00 +0000

Description

FizzBuzz problem was made popular by Joel Grus’ post on tensorflow. This article shows how to model fizzbuzz using MXNet. This article shows how to use the basic building blocks of MXNet (autograd etc.) to build a network from scratch. It then goes to show how to use higher level APIs in MXNet to achieve the same.

Read more..