ML Engineer · 5+ years in NLP & LLMs
# trained from scratch. no pre-trained weights.
Initialized with a Mechanical Engineering degree, pre-trained on curiosity, and fine-tuned by an unreasonable number of late nights — 5+ years of gradient descent through NLP, LLMs, and generative AI. Currently inference-serving at Apple. I write about ML here so you don't have to train from scratch too.
Apple — multi-agentic systems & LLM research
Notebooks GM, Competitions Master
Dynamic intent detection system
ME degree → ML through sheer will
model.fit(life, epochs=∞, lr=persistence)
Each milestone is a neuron. Experiences flow forward. Lessons backpropagate. The loss is still decreasing.
training_loss.plot()
“You don’t need a low initial loss. You need a good learning rate and the patience to keep training.”
— the philosophy that took me from Mechanical Engineering to Apple
latest
The shift from absolute to relative positional encoding was one of the most important architectural changes in modern LLMs. This post covers T5's relative bias, ALiBi's simplicity, and RoPE's elegant rotation — the method that won and powers every major LLM today.
Without positional encoding, a transformer treats 'the cat sat on the mat' and 'mat the on sat cat the' as identical. This post explains how sinusoidal and learned positional encodings give transformers a sense of order, and why the field moved beyond them.
Tokenization is the invisible foundation of every LLM. This post traces how it evolved from simple word splitting to BPE, WordPiece, Unigram, and SentencePiece — and why the choice of tokenizer shapes everything downstream.
Explore how DeBERTa revolutionizes NLP with its innovative Disentangled Attention Mechanism and Enhanced Mask Decoder, surpassing previous models like BERT and RoBERTa in performance and setting a new benchmark in the field
Watch my conversations with people who shaped the Indian ML community: