Welcome to DeeperThoughts – A blog that bridges the gap between code and cognition.

Hi, I’m Rohith. I’m a Research Scientist at the Centre for Development of Telematics (Govt. of India), with a keen interest in AI and deep learning—especially multimodal systems. This blog is where I share thoughts, projects, and ideas that help me make sense of the what I read about or learnt (and hopefully make it a little easier for you too).

If something sparks a thought, I’d love to hear from you in the comments. And if you’d like updates whenever I post, feel free to Subscribe to the newsletter and get the latest straight to your inbox 📬.

Showing 10 of 10 posts

Published on
August 11, 2025
The AI Video Revolution with Veo3 and Latent Diffusion
Generative-AI Video-Generation veo3 Developer-Tools Computer-Vision-
AI video generation is transforming the way we create visual content. Models like Veo 3 can turn simple text prompts into lifelike, dynamic videos, opening new possibilities for storytelling, education, and entertainment. But how exactly do these AI models generate realistic motion and visuals from just a few lines of text?
Read more →
Published on
August 3, 2025
How do LLMs solve Math Equations?
LLMs Generative-AI Interpretability Optimization AI-efficiency
Large Language Models like GPT-4 and Gemini werent built as calculators, but can now solve many math problems. They work by recognizing and reproducing patterns, not by “understanding” math in a human sense, so errors and hallucinations still occur. This blog explores how LLMs handle math and the challenges they face.
Read more →
Published on
July 26, 2025
Deploying AI in Agriculture: How we built a Semi-Supervised ViT pipeline for ICAR
ViT Semi-Supervised-Learning Computer-Vision Automation MLOPs ONNX
What if deep learning could empower farmers to diagnose crop diseases in the field using nothing more than a smartphone? This article explores the implementation of our project, built under a joint ICAR–NITK collaboration, and how Semi-Supervised learning, Vision Transformers, and ONNX deployment was used to create a diagnostic model that achieved 95.7% accuracy—ready for real-world use in the field.
Read more →
Published on
July 20, 2025
How to Set Up an MCP Server for Multi-Agent AI Systems
AI-Agent MCP Automation Developer-Tools Generative-AI
Learn how to set up a lightweight MCP (Model Context Protocol) server to enable context sharing across AI agents like ChatGPT. This article walks through the core concepts and includes practical code examples to help you build context-aware, multi-agent AI systems.
Read more →
Published on
July 13, 2025
LLM Vulnerabilities: From Hidden Threats to Secure Deployments
LLMs Optimization MLOps AI-Security Developer-Tools
Large Language Models (LLMs) are revolutionizing AI applications across industries. But with their power comes a growing list of security risks—prompt injection, data leakage, and model manipulation. As these threats evolve, one question remains: are we doing enough to protect our AI systems against these threats, and what more can we do?
Read more →
Published on
July 6, 2025
Visual Perception through Vision Transformers
Generative-AI ViT Computer-Vision Attention Interpretability
Vision Transformers (ViTs) have transformed the way machines process and understand images, replacing convolutions with self-attention. At the heart of this shift lies the question: how do these models “see”? This article explores the internal workings of ViTs, focusing on attention heads, patch embeddings, and interpretability techniques that help us decode how these models perceive the visual world.
Read more →
Published on
June 30, 2025
Is Cursor the End for Programmers? The Rise of Code-LLMs and Vibe Coding
LLMs Generative-AI Developer-Tools NLP
As Cursor transforms how developers write code, we are witnessing the emergence of "Vibe Coding"—a new paradigm where programmers work through intuition and intent rather than explicit syntax. But how are these CodeLLMs different from traditional LLM architectures, and are we really losing the essence of what made us programmers in the first place?
Read more →
Published on
June 23, 2025
Is Bigger Always Better? Rethinking AI Scaling Laws through Small Language Models
LLMs scaling-laws Optimization Generative-AI ai-efficiency
Exploring how 1B-parameter small language models can outperform 405B giants through clever compute allocation, test-time scaling, and innovative training strategies. Is bigger really always better in the AI world?
Read more →
Published on
June 16, 2025
Why GANs Fool Us So Well—Until They Fail
GAN CNNs Computer-Vision Generative-AI
Ever came across reels on Instagram that show Queen Elizabeth breakdancing and wondered how they look so uncannily real? In a world where AI can create art, faces, and even deepfakes from scratch, GANs stand at the center. But how does AI really generate images that even humans can’t tell are fake?
Read more →
Published on
June 10, 2025
Beyond Attention—A Dive into the Interpretability of Self Attention Heads in modern LLMs
LLMs Attention NLP Transformers Generative-AI Interpretability
New capabilities of Large Language Models like GPT, LLaMA, Claude, and Grok are emerging almost every day, and at the heart of these powerful models lies the multi-head self-attention mechanism—the secret behind their impressive reasoning and language skills. This article takes a closer look at how these attention heads work together under the hood to create the tools we rely on and admire today.
Read more →

Subscribe to the newsletter

Deeper Thoughts – AI and Deep Learning Blog

Welcome to DeeperThoughts – A blog that bridges the gap between code and cognition.

Latest Blog Posts

Newsletter Subscription