Dr. Joy Bose: Projects

Selected projects across Ericsson, Microsoft, Samsung, and PhD research. For full publication list see Research or Google Scholar.

Ericsson Global (2020–present)

Knowledge Graph + MCP Agentic System for Code Understanding

2023–present

Designed and deployed a knowledge graph and LLM-based agentic solution using FalkorDB and MCP for intelligent code understanding and log querying via OpenSearch. Adopted by 1,000+ engineers across multiple BCSS teams.

LLM-Based Automated Java Code Review

2022–2024

Built an automated code review pipeline using open-source Llama models, adopted by ~400 engineers across the BCSS organisation. Estimated saving of ~20 minutes per review cycle. Paper published at ICSME 2025.

Publications: Ramesh S., Bose J. et al. Automated Code Review Using Large Language Models at Ericsson. ICSME 2025.

NLP-to-SQL System with RAG for Telecom Data Access

2021–2022

Built a retrieval-augmented generation pipeline enabling natural language querying of complex telecom datasets, eliminating manual query bottlenecks for multiple teams.

Anomaly Detection Using Hierarchical Temporal Memory

2021–2023

Developed a hybrid real-time data drift and anomaly detection system combining HTM and statistical tests. Published in IJMEMS journal (2025) and presented at ORSI ICBAI 2023.

Publications: Bandyopadhyay S., Bose J. et al. A Hybrid Framework for Real-Time Data Drift and Anomaly Identification. IJMEMS 2025.

ML for Telecom Energy Efficiency and Trouble Ticket Prediction

2020–2022

Applied ML to identify inefficient radios for energy savings and predict trouble ticket resolutions. Published at COMSNETS 2022 and 2024.

Financial Variance Categorisation for Telecom BSS

2020–2021

Classified financial variance data for telecom providers using ML, reducing manual root cause analysis time. Published on Ericsson tech blog 2021. One patent filed (EPO/WIPO).

Microsoft Edge Team (2018–2019)

System to extract relevant content on a webpage for reader mode

Aug 2018 – Apr 2019

Developed an ML application to extract content from webpages based on relevance and remove ads or boilerplate. Used Bing's Virtual DOM extractor. 1 US patent granted (US12591731B2).

Publications: Extraction of relevant images for boilerplate removal in web browsers. INDICON 2019.

Machine learning to predict field labels in an autofill system

Apr – Dec 2019

Built a client-side neural network model to predict field labels for the autofill feature in Edge. Model saved in ONNX format using WinML.

Samsung R&D (2011–2018)

Modelling attention in web browsers using EEG

Jan 2014 – Jan 2016

Explored uses of attention as detected by portable EEG sensors (NeuroSky MindWave) while browsing the web and performing activities on smartphones.

Publications: A Hands Free Browser Using EEG and Voice Inputs (IJCNN 2015); EEG Based Detection of Area of Interest in a Web Page (ICACCI 2015); Attention Sensitive Web Browsing (ACM Compute 2016).

Awards: Best Paper Award, Track 4, INDICON 2015.

Bias detection in web browsers

Jan 2016 – Jun 2018

Investigated methods to detect bias in news articles being read on the mobile web browser in real time.

Publications: Bias Based Navigation for News Articles and Media (NLDB 2016); Bias Aware Web Browsing (ICMLA 2018).

Calendar event classification

Jun 2017 – Aug 2018

Built ML models to classify calendar data into predefined categories in both Korean and English using an ensemble of techniques. 94% accuracy.

Publications: A Personalized Health Recommendation System based on Smartphone Calendar Events (ICOST 2018).

Increasing the click rate of web notifications

Jan 2015 – Oct 2016

Built a model predicting which notification categories users are likely to click based on past behaviour and time of day.

Publications: Secure Web Push System (COMSNETS 2016); Intelligent Web Push Architecture (IEEE ICWS 2016).

Patent: US Patent US20160014057A1 — Method and system for providing dynamically customized web push messages.

Responsive and adaptive web browsing

Jun 2013

Investigated systems to make the web browser adaptive and responsive to user needs including head position, environmental conditions, and context.

Publications: Contextual Adaptive User Interface For Android Devices (INDICON 2013); Responsive, Adaptive and User Personalized Rendering on Mobile Browsers (SEWAD 2014).

Encrypting sensor data based on user gestures

Jun 2012 – Jun 2013

Methods to encrypt and decrypt data in mobile devices based on user gestures.

Publications: Encryption in mobile devices using sensors (SAS 2013).

Humanitarian engineering

Jun 2012

Investigated methods to make technology accessible to people in rural or remote areas.

Publications: A solution for a mobile computing device for illiterate users in rural areas (INDICON 2012); A kiosk based model for employment generation in rural areas (GHTC-SAS 2014).

Embibe (2020)

Determining optimal nudges for improving student scores

Jan – Jun 2020

Used optimisation algorithms to identify parameters where students could improve test scores, generating personalised nudge messages.

Publications: Auto generation of diagnostic assessments and their quality evaluation (EDM 2020).

PhD Research — University of Manchester (2002–2007)

Spiking neural sequence machine

2002–2007

Built software models of scalable spiking neural architectures for sequence learning using rank-order codes. 7 publications including 1 IEEE TNN journal paper co-authored with Steve Furber.

See Research page for full description and publications.