I turn data into decisions.
Building machine learning systems
from first principles to useful interfaces.
A few problems.
A lot of possibilities.
Predicting taxi trip duration with multiple linear regression, geospatial features, and time attributes. A Streamlit dashboard brings trip simulation and demand analytics together.
A handwritten digit classifier built with TensorFlow and Keras. Dense layers, ReLU activations, and a Softmax output turn pixel inputs into ten digit classes.
A multi-feature real estate ML system with vectorized linear regression, gradient descent, and feature normalization. Explore predictions through an interactive Streamlit interface.
Don’t just read about it.
Watch a model learn.
Gradient descent in action. Change the learning rate, add points to the chart, and watch a linear model fit the data.
An educational experiment with synthetic data—not a benchmark of my project models.
Each dot is a sample. The orange line predicts y from x. Each training step adjusts its slope and intercept to reduce the mean squared error. A smaller loss means a closer fit to these points.
Move two inputs through a tiny neural network. Every node value and output probability updates instantly using a real forward pass.
A 2 → 3 → 1 network with fixed illustrative weights and sigmoid activations. This is an explanatory demo, not a trained classifier or project performance result.
σ(z) = 1 / (1 + e⁻ᶻ)
h₁ = σ(3x₁ + 2x₂ − 2)
h₂ = σ(−2x₁ + 3x₂ − 0.5)
h₃ = σ(2x₁ − 3x₂ + 0.5)
output = σ(2h₁ − h₂ + 1.5h₃ − 1)
I’m Varun, an AI/ML Engineer at Futura Edtech, with a computer science background and a curiosity for how intelligent systems work.
My projects trace a path from implementing gradient descent by hand to building neural networks and interactive ML applications. I’m interested in both sides of the equation: the model underneath and the experience it makes possible.
Understand what happens beneath the library call.
Connect a model to a problem people can understand.
Build, inspect, and make the next version better.
Python · NumPy · Pandas · Scikit-learn · TensorFlow · Keras · Streamlit
Good intuition.
Built through repetition.
Algorithmic problem solving on LeetCode.
Building precision, one solution at a time.
Profile snapshot · October 2026 · supplied LeetCode profile
Interesting problems. Thoughtful teams.
There’s always a good reason to connect.