Data scientist and ML engineer in melbourne

predictive models, deep learning, computer vision and analysis, taken past the notebook and into the workflow where people make decisions.

A model is only useful once it is in front of the people who make decisions with it. My background is in both halves: the statistics and machine learning (a BS in Data Science from IIT Madras and a Master of Data Science at the University of Melbourne), and the engineering that puts a model into a product, a dashboard or a clinical workflow.

At IIT Ropar I built a desktop platform that runs the AI in Medicine Group’s deep-learning models on 2D and 3D medical images for clinicians at PGIMER Chandigarh, a hospital with a daily footfall of over ten thousand people, and lets them flag and re-annotate poor outputs for the next round of fine-tuning.

what i build

  • Predictive modellingForecasting, classification and pricing models on tabular data, with honest validation and features you can explain.
  • Deep learning and computer visionImage classification and segmentation, including medical imaging with MONAI and nnU-Net, in PyTorch.
  • NLP and text analyticsClassification, sentiment, extraction and search over text, with classical models or LLMs as the task demands.
  • Analysis and dashboardsStatistical analysis and dashboards in Kibana or a custom web app, so the numbers are visible to the people who need them.
  • Model deploymentServing models behind an API or inside a desktop or web app, with feedback loops for retraining.

work i have shipped

AI in Medicine Group, IIT Ropar, with PGIMER Chandigarh

Clinical deep-learning deployment platform

Inference on histopathology, FISH, MR and CT images, with poor outputs flagged and re-annotated to feed fine-tuning.

University of Melbourne

Distributed social sentiment analysis

Mastodon and Bluesky streams processed with local LLMs into live Kibana dashboards on a six-node Kubernetes cluster.

IIT Madras

Taxi fare prediction

End-to-end pipeline with cleaning, outlier detection and feature engineering, comparing Linear Regression, Random Forest and XGBoost.

all projects →

tools

PythonPyTorchScikit-LearnXGBoostPandasNumPyMONAInnU-NetMatplotlibSQLElasticsearchKibana

how it works

  1. A short callWe talk through the problem, who it is for and what done looks like. No preparation needed.
  2. A written scopeA short document with milestones, what is in and out of scope, and a quote or an hours-per-week arrangement.
  3. Build in the openYou see working software early and regularly, with tests and deployment set up from the first commit.
  4. HandoverDocumentation, access to every account and a walkthrough, so your team can run it without me.

how i think about it

questions

We have data but no clear question. Can you help?

Yes. The first step is usually an exploratory analysis that shows what the data can and cannot answer, before any model is built.

Deep learning or classical machine learning?

For most tabular business data, gradient-boosted trees and well-chosen features beat deep learning and are easier to explain. Deep learning is the right tool for images, audio and large amounts of text.

How do we know the model will work in practice?

By validating on data that resembles what the model will see later, measuring the metric that matters to the business, and monitoring it after launch.

also hired as

Data Scientist · Machine Learning Engineer · MLOps Engineer · Deep Learning Engineer · Computer Vision Engineer · Medical Imaging AI Engineer · NLP Engineer · Data Analyst · Data Engineer · Dashboard Developer · Statistical Modeller · Research Engineer

other services

have a project in mind?

send a few lines about what you need and when. the first call is free and there is no obligation after it.