Active · ML Engineering
March 2026 — Bengaluru, India

Aditya ND

@wimpykidaditya

Building intelligent systems from the ground up.
Student. ML engineer in progress.
Depth over breadth, always.

morning: math & theory evening: code & build writing: ML from Scratch series
p. 01

About

I'm a CS (AI & ML) student based in Bengaluru, working through a structured roadmap to become a core ML engineer with MLOps capability.

My approach is depth over breadth — every project is deliberately ordered, building compounding understanding from first principles to production systems.

LightGBM NumPy PyTorch FastAPI MIT 18.06 Backprop
p. 02

Projects

Flagship
01
AQI Construction Impact Predictor

Dual-model LightGBM system — regressor for AQI delta, classifier for spike risk. Does construction intensity explain AQI variability beyond weather? R² improved +26.6%, spike recall +27%. Served via FastAPI + Docker.

LightGBM FastAPI Docker AWS Research
Complete
02
Neural Network from Scratch

Generalized neural network via configurable layer_dims, ReLU hidden + sigmoid output, full backpropagation with L2 regularization. 96.67% accuracy on make_moons. NumPy only — no PyTorch, no TensorFlow. Just the math, made to run.

NumPy Calculus Backprop From Scratch
Open Source
03
AI-Scientist-V2 Contribution

Open-source contribution to AI-Scientist-V2 — research automation tooling for LLM-driven scientific paper generation pipelines.

Open Source LLM Python
p. 03

Technical Writing

ML from Scratch — a series building every core algorithm by hand, from the math up. No black boxes. Published on Hashnode, cross-posted to LinkedIn and X.

All posts on Hashnode →
p. 04

Skills

ML / AI
  • LightGBM · XGBoost
  • Scikit-learn
  • PyTorch (learning)
  • Anomaly Detection
  • K-Means Clustering
  • Regression · Classification
Engineering
  • Python · NumPy · Pandas
  • FastAPI
  • Docker
  • Git · Linux
  • AWS (learning)
  • REST APIs
Mathematics
  • Linear Algebra (MIT 18.06)
  • SVD · Orthogonality
  • Calculus · Gradients
  • Probability & Stats
  • Information Theory
p. 05
March 2026, Bengaluru
Let's build something.

Open to research collaborations
and problems worth losing sleep over.
Reach out — I respond.

— Aditya ND
p. 06