An interactive web app that utilises current game, past performance, and motion capture data to deeply analyse team and player performance, and subtler (and revealing!) insights such as field strategies, team cohesion, leader-follower relationships and so on.

A fully marker-less & real-time computer vision solution that serves as your personalised training assistant for your weekly dose of squats. Data ingestion was platform-agnostic, using a web app, and mediapipe ran on a single cloud node.

A facial recognition driven identity verification module for staff and/or visitor management.

A computational model to compute synchronization between groups of people, in small-group settings such as orchestras, football players, child-caregiver, and so on. Data extracted using human pose estimation algorithms.

A computational model developed to measure the direction of information flow between groups of two. Data extracted using human pose estimation algorithms, and computations made using causality analysis techniques.

A quickly tinkered and vanilla code that provides an easy-to-use interface for performing Granger causality tests on time series data. Built it initially for myself, and then had a colleague ask for it to analyze whether certain metrics help in deciding whether a goal has occurred in football matches or not. It solved a problem.

A pythonic implementation of the Multi-Event-Class Synchronization (MECS) algorithm, inspired by the paper The Multi-Event-Class Synchronization (MECS) Algorithm. This library provides tools to compute synchronization metrics for multi-event-class time series data, making it both intuitive and adaptable for all research and industry folks.

This Python script retrieves the metadata for a paper using the Crossref API, constructs a citation string in APA style, and returns it. Solved a problem for a colleague, and later for myself, to download metadata for a client in the pharma sector.

A scalable system that is now being used by 11 million users worldwide. Built the entire infrastructure from scratch, utilising state of the art data processing techniques, optimised+quantized proprietary ML models. Capable of performing real-time and batch inference. Multiple businesses have found deep value with increased average customer basket size, increased user retention, higher page/visits, lowering bounce rates, and increased average CLV.