devansh’s blog – Devansh's Blog

Unfiltered Lessons from Leading TEDx

TEDx

This is a collection of the lessons I learned about leadership, culture, and the surprising truths you uncover when you’re responsible for turning a shared dream into a reality.

Jul 22, 2025

Devansh Lodha

Exploring Vision-Language Models for Satellite-Based Brick Kiln Detection

computer vision

object detection

Brick kilns are a major, often hidden, source of air pollution. We explored how modern Vision-Language Models can be taught to find them from space, discovering that these new models can outperform traditional models with a tiny fraction of the training data.

Jun 16, 2025

Devansh Lodha

Power Plants, Pollution, and Pixels

computer vision

data analysis

code

We’ll analyze global power plants and use Google Earth Engine to build a machine learning model that can see pollution from space. We collect and use features from multiple satellites to accurately detect when a power plant is active.

Jun 9, 2025

Devansh Lodha

Vision-Language Models Tutorial

code

computer vision

An overview of everything leading up to the current state of Vision-Language Models, including their architecture, training, and applications. This tutorial was conducted at the ACM India Summer School 2025.

Jun 4, 2025

Devansh Lodha, Shataxi Dubey

The Vision Transformer (ViT)

bare-bones-ml

code

We’ll dive into the Vision Transformer (ViT), a Transformer architecture that can achieve state-of-the-art results in image recognition despite not having inductive biases like CNNs.

Jun 1, 2025

Devansh Lodha

Training Our Transformer

bare-bones-ml

code

We’ll teach our Transformer model to reverse sequences of numbers, a task that requires understanding relationships between input and output positions.

May 29, 2025

Devansh Lodha

Assembling the Full Transformer

bare-bones-ml

code

We’ll be putting together all the components we’ve built in the previous posts to assemble a complete Transformer model, ready for training and inference.

May 28, 2025

Devansh Lodha

The Attention Mechanism

bare-bones-ml

code

We’ll be implementing the attention mechanism, a powerful technique that allows models to focus on relevant parts of the input sequence at each step.

May 27, 2025

Devansh Lodha

LSTMs - Giving Your Network a Long-Term Memory

bare-bones-ml

code

We’ll be solving the vanishing gradient problem by building a Long Short-Term Memory (LSTM) network from scratch, implementing the forward and backward passes, and training it on a sequence prediction task.

May 26, 2025

Devansh Lodha

Building a Multi-Layer Perceptron for Classification

bare-bones-ml

code

We’ll be stacking multiple layers and tackling a real-world binary classification task.

May 25, 2025

Devansh Lodha

Recurrent Neural Networks from Scratch

bare-bones-ml

code

We’ll build a recurrent neural network from scratch, implementing the forward and backward passes, and training it on a sequence prediction task.

May 25, 2025

Devansh Lodha

From Tensors to a Trainable Neural Network

bare-bones-ml

code

We’ll use our autograd engine to build an initial deep learning library. We’ll move up one level of abstraction to create reusable layers and an optimizer!

May 24, 2025

Devansh Lodha