Foundations
Floating point, vectors, matrices, neural networks, embeddings.
Machine Learning · LLM Systems · AI Infrastructure
I build and write about large-scale machine learning systems.
I’m Pratheek Vadla, a machine learning engineer interested in LLMs, inference systems, recommendation systems, GPUs, and the infrastructure required to run AI reliably in production.
Featured series
LLMs, from first principles
A structured journey from numbers and neural networks to transformers, inference engines, GPUs, and production serving.
Coming soon
01 — How computers represent numbers
Before FP16, BF16, quantization, and tensor cores, we need to understand how computers represent numbers in the first place.
Transformers & LLMs
Attention, tokenization, pretraining, fine-tuning, alignment.
LLM Inference
KV cache, batching, PagedAttention, quantization, parallelism.
Production Systems
GPUs, Kubernetes, model serving, observability, cost and scale.
Research
Applied research in recommendation and LLM systems.
My interests sit at the intersection of machine learning systems, recommendation, retrieval, and large language models.
Research page →Projects
Experiments, systems, and open-source work.
This section will collect hands-on work around inference, retrieval, distributed systems, and model serving.
Projects page →About
Machine learning engineer based in the San Francisco Bay Area.
I work on production machine learning systems and write to understand difficult ideas from first principles. This site is where I document what I learn and build.