I am a Machine Learning Engineer with 7 years of experience in LLM post-training and computer vision research. I have previously built tools used by ML practitioners around the world at Weights & Biases, led a team of applied scientists on agentic AI post-training at Eternal for India’s biggest mega-apps ranging from food delivery, to quick commerce, and AI-powered wearables.
My Career Timeline
| Position | Organization | Duration |
|---|---|---|
| Lead Machine Learning Researcher | Eternal | March 2026 - September 2026 |
| Machine Learning Researcher | AthenaAgent | July 2025 - February 2026 |
| Machine Learning Engineer | Roboflow | April 2025 - July 2025 |
| Machine Learning Engineer | Weights & Biases | January 2022 - March 2025 |
| Staff Machine Learning Engineer | Ignitarium | July 2021 - December 2021 |
| Software Development Engineer | IBM | January 2020 - July 2021 |
| Software Development Engineer | HighRadius | June 2019 - December 2019 |
| Failed entrepreneur and Machine Learning Researcher | DeepWrex Technologies | November 2018 - December 2019 |
Additionally, I’ve been a Google Developer Expert for Machine Learning (JAX) since Jun 2022.
Major Projects
- Inksphere: Transform books into immersive reading experiences (Demo).
- Psynapse: A cross-platform node-based UI editor for building AI workflows using Python.
- Trackers: Trackers is a unified library for object tracking featuring clean room re-implementations of leading multi-object tracking algorithms. I am one of the core maintainers actively implementing SoTA multi-object trackering techniques, Re-ID models, training, and fine-tuning pipelines. Check out the docs for more information.
- WandB Models: WandB Models is the AI developer platform used to train and fine-tune models, and manage models from experimentation to production. I have contributed mojor integrations of WandB Models with open-source ML libraries like Hugging Face Diffusers, Hugging Face AutoTrain, Keras, MMEngine, PyTorch Geometric, Ultralytics, YOLOv5, MONAI, etc. I have also authored numerous technical reports and developed experimental tooling for ML practitioners.
- WandB Weave: WandB Weave is a toolkit for developing Generative AI applications, built by Weights & Biases. I have contributed mojor integrations of Weave with LLM SDKs like Groq, Hugging Face Hub, Google; and LLMOps frameworks like DSPy, Instructor, and SmolAgents.
- Hemm: Holistic Evaluation of Multi-modal Generative Models: Hemm is a library for performing comprehensive benchmark of text-to-image diffusion models on image quality and prompt comprehension integrated with Weights & Biases and Weave.I am currently actively working on this project. Check out the docs for more information.
- Weights & Biases Addons: Weights & Biases Addons is a repository that provides of integrations and utilities that will supercharge your Weights & Biases workflows. Its a repositpry built and maintained by WandB users for WandB users. The library hosts experimental utilities and integrations built using Weights & Biases. I am currently actively working on this project. Check out the docs for more information.
- Restorers: Restorers is a library provide out-of-the-box TensorFlow implementations of SoTA image and video restoration models for tasks such as low-light enhancement, denoising, deblurring, super-resolution, etc. You can read more about it in this WandB report.
Deep Learning examples published on keras.io
- Multiclass semantic segmentation using DeepLabV3+
- Low-light image enhancement using MIRNet
- Zero-DCE for low-light image enhancement
- Large-scale multi-label text classification in colaboration with Sayak Paul
- Point cloud segmentation with PointNet in colaboration with Sayak Paul
- GauGAN for conditional image generation in colaboration with Sayak Paul
Other Interesting Projects
- Syntia: Syntia brings modern IDE features to your terminal. Built with Textual, Syntia combines the power of a traditional code editor with the convenience of an integrated terminal environment.
- Radium: A small and lightweight Ray Tracing Engine written in C++ that runs on the CPU using shared-memory multiprocessing.
- Colorization using Optimization: Python and C++ implementations of a user-guided image/video colorization technique as proposed by the paper Colorization Using Optimization. The algorithm is based on a simple premise; neighboring pixels in space-time that have similar intensities should have similar colors. This premise is formalized using a quadratic cost function that obtains an optimization problem that can be solved efficiently using standard techniques. While using this alogorithm, an artist only needs to annotate the image with a few color scribbles or visual clues, and the indicated colors are automatically propagated in both space and time to produce a fully colorized image or sequence. The annotation can be done using any drawing tool such as JSPaint or Gimp.
- Deep Deterministic Policy Gradients: Pytorch implementation of the Deep Deterministic Policy Gradients Algorithm for Continuous Control as described by the paper Continuous control with deep reinforcement learning.
- Twin Delayed DDGP: Pytorch Implementation of Twin Delayed Deep Deterministic Policy Gradients Algorithm for Continuous Control as described by the paper Addressing Function Approximation Error in Actor-Critic Methods.
- Arxiv2Kindle: Arxiv2Kindle is a simple script written in python that converts LaTeX source downloaded from Arxiv and recompiles it to better fit a reading device (such as a Kindle).
- Manga Scraper: A a python package that downloads Manga into chapterwise PDF files or a single PDF file from various sources. It basically adds a post-processing layer on top of the basic functionality by
mangadl-bashcreated by Akianonymus in order to convert the downloaded manga into chapter-wise PDF files or a single giant PDF file.