I'm a developer based in India, working through a structured, self-directed program to build modern AI/ML systems — autograd engines, neural networks, optimizers, and eventually transformers — entirely from scratch in Python. No shortcuts through high-level libraries doing the thinking for me.
The goal isn't a working model, it's the derivation — understanding linear algebra, multivariate calculus, probability, and information theory well enough that I could have built every piece myself, not just called .fit() on it.
I also carry a solid full-stack web & mobile background, and still take on freelance work on the side.
A day-by-day path from raw math to a working, trained language model — and beyond. Order here isn't decorative, each stage is a prerequisite for the next.
| ✅ | Math Foundations & Autograd | Hand-derived SVD, eigendecomposition, PCA · Jacobians & the chain rule · a working autograd engine · a Neuron → Layer → MLP library on top of it · a custom Adam optimizer with a deliberate tweak of my own |
| 🔄 | Probability, Info Theory & Classical ML | Naive Bayes · Decision Trees · Random Forests · SVMs · K-Means — every one implemented from scratch and validated against scikit-learn |
| 🎯 | Deep Learning & Transformers | PyTorch · CNNs & RNNs · Transformers · a GPT built from scratch |
| 📍 | Alignment & Beyond | Fine-tuning (LoRA, DPO) · generative models · reinforcement learning |
📂 The full day-by-day math, code, and write-ups live in AGI_Research — if you want the actual derivations and implementations rather than the summary above, that's the place to look.
AI/ML & Research — the main focus right now
Full-Stack — prior work, still freelancing
Available part-time for:
- Full-stack web apps
- Flutter mobile apps (iOS/Android)
- Backend APIs
- Python-based automation & data scripts
Clean code, honest timelines — I actually respond to messages.
"I don't just use the tools — I build them, so I actually understand how they work."