Hey, I'm Jyotinder
I'm a software engineer interested in low-level software systems and largely work with C++, Go, and Python. I'm currently working on Keras to make machine learning easier for humans. Previously built query engines at Google Ads.
Selected work
09Keras
Deep learning for humans: a multi-backend framework for JAX, TensorFlow and PyTorch. Home of the Keras Quantization API and the model optimization charter.
KerasHub
Pretrained models for Keras 3: multi-framework implementations of popular architectures with checkpoints on Kaggle. I built its end-to-end automated release pipeline.
Qwix
A quantization library for JAX: quantization-aware training and post-training quantization for XLA and LiteRT targets. I built its benchmark framework and low-bit techniques such as INT4, GPTQ and AWQ.
Kinetic
Run Keras and JAX workloads on cloud TPUs and GPUs with a single decorator, with no infrastructure to manage.
DiceDB
An open-source, low-latency key/value engine built on Valkey with query subscriptions: a multi-threaded core, reactive subscriptions, an observability stack and client SDKs.
Apache Ozone
A scalable, redundant, distributed object store for Hadoop and analytics workloads.
dropdb
A from-scratch relational database in Go: disk and memory management, transactions, query planning, and Go's standard sql interface.
goLSM
A write-optimized embedded key-value store built on a log-structured merge tree in Go; ~10 µs reads and writes under the Yahoo Cloud Serving Benchmark.
Tok
A programming language with its own bytecode virtual machine in C: dynamic typing, garbage collection, closures, classes and inheritance, plus a remote execution engine so it runs in the browser.