Sitemap

Ram Balachandran - Chief Data Scientist building AI for physical systems

Git hooks are a simple, easy way to do small local scripting to improve code quality and readability. Using consistent git hooks across projects can reduce the load on CI/CD process which can focuse more on unit and integration testing.

Any data scientist or machine learning interested in developing production deployable models need to understand Docker. Docker is one of the most important tools for DevOps that can help to create and maintain a consistent enviroment across development, staging and production. In particular docker-compose can be used to configure and interface applications using multiple docker containers.

Provisioning an EC2 spot instance by hand every day eats productive time. A small boto3 script plus crontab automates the whole routine.

Extracting text from images using Optical Character Recognition (OCR) has enormous applications across almost all industries. A robust, scalable OCR solution is an integral part to the digital transformation journey of every organization. These solutions can help an organization to obtain a significant edge over its competitors. For example, OCR solutions can

1. Help to reduce the time required in processing paper forms.
2. They can digitize information from old important records that can be easily looked up for reference or can be analyzed to build insightful data models.
3. Finally, OCR can be used along with images and videos to extract important textual information.

I have used many IDE environments over the past years ranging from emacs, pycharm, eclipse among others. Emacs used to be my favorite editor since I was able to customize it to my needs. For example, one of the things I loved about Emacs was its seamless ability to work on code in remote servers through the Tramp package. However, in a corporate career, you are mostly destined to work in local Windows environment and customizing Emacs to work with windows was becoming too much of a pain.

Computer vision and signal processing that automate industrial inspection, from detecting clay seams in potash mines to finding cracks in wastewater pipes.

Parcel-level automated valuations with up to 90% accuracy for 200,000+ South Australian properties, built on continuously refreshed sales, cadastral, zoning and auction data.

Physics-constrained machine learning and multitask graph neural networks for designing batteries, supercapacitors and fuel-cell materials from small datasets.

Founding member of Dragonfly, built with Jacobs Water, which turns raw sewer CCTV into prioritised maintenance actions with >95% PACP coding accuracy.

A multi-modal ingestion pipeline (parsers, speech-to-text, embeddings, vector search, NL2SQL) that lets teams query documents, calls and databases with every answer citing governed data.

An AWS-native platform that turns >100 Hz pressure data into transient-risk views within ~15 minutes, helping utilities find and remove the water-hammer sources that shorten pipe life.

Replaced spreadsheet-only trade planning for Endeavour Group with a guided AI workspace that models uplift, elasticity and cannibalisation at store–SKU–week depth.

A low-cost smart building energy management prototype combining Raspberry Pi sensing, computer vision and a physics-based R-C thermal model to adapt HVAC setpoints.