CareStack

Jul 2015 – Aug 2023 · Trivandrum, India

All-in-one dental practice-management platform serving hundreds of clinics.

CareStack is the dental practice-management platform I worked across for eight years, joining as a junior developer and resigning as Senior Staff Engineer. It brings scheduling, clinical charting, billing and front-office operations into one system, serving hundreds of clinics.

I owned software architecture and technical design across mission-critical, high-throughput workflows, and drove database design and deep SQL performance optimisation on Azure SQL Server and Cosmos DB.

CareRevenue

CareStack · led end to end

Cloud-native revenue-cycle-management platform for seamless dental RCM services, led and architected end to end.

CareRevenue is the revenue-cycle-management platform I led from the front: team lead, architect and cloud owner in one. Cloud-native from the ground up on Azure, it handles high-throughput, mission-critical billing workflows for seamless dental RCM services.

The work spanned software architecture and technical design, database design and deep SQL performance optimisation, and cloud and CI/CD engineering, while leading the cross-functional team delivering it.

ReproAlign

ReproAlign (co-founded)

The world’s first integrated fertility intelligence system, built from zero as co-founder.

ReproAlign is the reproductive-health AI company I co-founded as Head of Engineering & AI. We are building the world’s first integrated fertility intelligence system.

I lead product architecture and AI integration: intelligent data pipelines, predictive analytics and recommendation systems, and established the engineering practice, CI/CD and cloud-native infrastructure from zero.

SOS Learning Platform

UK healthcare group (private)

A private LMS for a co-owned group of healthcare agencies and care homes.

SOS LMS is a private, multi-tenant learning management system for a co-owned group of healthcare agencies and care homes, not a public SaaS. It gives each agency its own space to train and certify staff against care-sector requirements.

The platform spans three roles: agencies enrol their staff as learners, learners work through course content and quizzes to earn certificates, and super admins operate the platform end to end, including billing and reporting.

Warehouse CAD to SQL

Consafe Logistics Group, Lund, Sweden · Feb – Jun 2025

Warehouse CAD layouts read by computer vision and written out as SQL database structures.

The Master’s thesis: Automated SQL Command Generation from Digital Diagrams Using Computer Vision Techniques, written with Consafe Logistics Group in Lund and graded 4 on a 3 to 5 scale.

A warehouse layout arrives as a CAD drawing and leaves as a relational database. YOLO models detect the layout elements in the drawing, and the pipeline turns what they find into SQL structures ready for a warehouse management system to consume.

The dataset was built by hand along the way: annotated in CVAT, augmented with Albumentations, and trained on an HPC cluster scheduled by Slurm.

Lead Time Prediction

Almax Mori · Università di Trento, Italy · Sep 2024 – Jan 2025

Predicting production delays for a global aluminium extrusion die manufacturer.

Built during the Erasmus+ exchange at Università di Trento, with Almax Mori: which orders in an aluminium extrusion die shop will run late, and by how much.

Gradient-boosted ensembles over the production history, delivered by an international, multidisciplinary team.

Depth vs RGB Detection

Viscando AB · Jönköping University · Mar – May 2024

The same detector trained on depth maps and on RGB, evaluated head to head.

A controlled comparison for Viscando AB: what is a depth image worth to a detector? YOLOv5 trained on RGB and on depth maps of the same scenes, then evaluated head to head.

KITTI stereo pairs converted to depth maps supplied the depth dataset, with OpenCV doing the conversion work.

Solar Production Forecasting

Wettersol AB · Jönköping University · Jan – Feb 2024

Daily solar energy production forecasts for a Swedish solar installer.

Daily production forecasting for Wettersol, a Swedish solar installer: how much energy will an installation produce tomorrow, and across the swing of a Nordic year.

LSTM networks against Random Forest baselines over production time series, with meteorological data integrated for the seasonal effects.

Claim Propensity Prediction

Dunstan AB · Jönköping University & Science Park · Oct – Dec 2023

A classification model for insurance claim propensity, and the data work under it.

The first applied AI project of the Master’s, for Dunstan AB: a classification model predicting claim propensity.

The client database was restructured to make the modelling possible, and the full methodology was documented for handover.

Say hello

Open to conversations about AI engineering, architecture and building teams.

The fastest channel is email, and the form below reaches the same inbox. LinkedIn and GitHub carry the rest of the record.