I build intelligent
systems from first principles.
AI/ML Engineer — from neural networks and backpropagation implemented by hand in NumPy, to production NLP and RAG pipelines. I understand what’s under the hood, then ship it.
Final-year CS student, already shipping ML.
I work across the full ML stack — classical algorithms (regression, SVM, ensemble methods), deep learning in PyTorch, and applied NLP: embeddings, transformers, LLM fine-tuning, and RAG with LangChain.
I pair that with real backend engineering — REST APIs, FastAPI, and full-stack delivery with React and PostgreSQL — so I can take a model from a notebook to a deployed product. I also bring ~2 years of professional full-stack experience solving production problems.
Off-screen: competitive cricket and table tennis, and ideas at the intersection of physics, philosophy, and technology.
Machine Learning
- Regression
- SVM
- Ensemble methods
- Feature engineering
- EDA
- Model evaluation
Deep Learning
- ANN
- CNN
- RNN
- PyTorch
- Optimization
NLP & LLMs
- Embeddings
- Transformers
- RAG architecture
- LangChain
- LLM fine-tuning
Backend & Delivery
- FastAPI
- REST APIs
- React
- MySQL
- PostgreSQL
- Git
MLOps & Deployment
- Docker
- CI/CD
- MLflow
- AWS
- Railway
Things I built to understand, then shipped.
Most of these are interactive explainers — I re-implement the idea from scratch, then make it something you can actually play with.
Backpropagation, Visualized
You scroll, the forward pass lights up layer by layer, the error appears at the output, then every ∂E/∂w flows backward through the chain rule — shown exactly where it lives in the network. No framework: vanilla JavaScript drawing the whole net as SVG, KaTeX typesetting the real equations, dark/light, fully offline.
↳ This site’s hero grew out of this project.
Interactive EDA Guide
Histogram? Boxplot? Scatter? An interactive guide to the whole EDA workflow — no walls of text.
Zyotis Guru
A production platform with an integrated RAG assistant — full-stack build, from data to deployed product.
HomeCare CRM
A CRM with data and ML-insight layers for a home-care service — working with real operational data.
↳ Private / login-gated — no public demo.
Two years of engineering, one clear direction.
A capability timeline rather than a résumé — what I can do, and how the ML focus grew out of solid full-stack foundations.
AI / ML Engineer
Building intelligent systems end to end — deep learning, applied NLP, and RAG — alongside interactive explainers that make the hard parts visible.
Machine Learning
Neural networks from scratch in NumPy, classical ML, deep learning in PyTorch, and production RAG with LangChain. Taking models from research into deployed products.
Full-stack Engineering
Shipping production backend applications and full-stack features, integrating dynamic front-ends, and solving real-world production problems. The most recent year has been fully ML-focused.
Notes on building things from scratch.
Short write-ups on the projects and the ideas behind them — mostly variations on one theme: understand it before you use the library.
Let’s build something meaningful.
Have a hard ML problem, an idea to build, or something to ship? The form reaches me directly.