AI Research Intern

Yorktown Heights, New York

Projects Highlights:

Design, develop, and deploy high-quality industrial AI agents, scenarios and custom evaluation protocols for the IBM agentic system (AssetOpsBench). Implemented embodied question answering platform and failure metadata knowledge graph. Built automated evaluation pipelines combining rule-based validation, LLM-based judgment and custom evaluation frameworks.

Neuro-symbolic AI Research Intern

Robert Bosch LLC - Pittsburgh, Pennsylvania

Project Highlights:

CausalPulse: Multi-Agent Root Cause Analysis Copilot for Smart Manufacturing

Designed and implemented a multi-agent Root Cause Analysis framework for smart manufacturing by integrating anomaly detection, causal discovery, and Root Cause Analysis agents into an interactive copilot interface. Benchmarked the system against existing causal copilots and RCA systems to demonstrate improved interpretability and adaptability.

Domain Knowledge-Guided Causal Discovery

Developed methods for integrating domain knowledge into causal discovery by designing and applying rule-based constraints to enhance learned causal graphs.

Observational and Control Data Fusion

Implemented a data fusion pipeline combining observational data and control data to improve causal graph learning and Root Cause Analysis for grinding-process use cases.

Real-World Manufacturing Evaluation

Evaluated the framework on real-world manufacturing datasets, including the Planar Sensor Element dataset and the Future Factories dataset. For the Planar Sensor Element use case, the framework supported causal analysis of robot configurations related to an approximately $15M/year scrap issue. The Future Factories dataset was used to validate the approach in an academic benchmark setting.

Graduate Research Assistant

The Artificial Intelligence Institute of UofSC (AIISC)

Projects Highlights:

Rare event prediction and anomaly prediction in manufacturing

Leading the Smart Manufacturing Project at AIISC, focusing on implementing supervised learning models, data enrichment techniques, 2D-based prediction, multimodal hybrid fusion techniques, and custom, compact neurosymbolic co-pilot system for rare event and anomaly prediction.

Causal Event Graph-Guided Language-based Spatiotemporal Question Answering

Implementation of an ontology and reasoning approach to handle physics-based spatiotemporal reasoning for CLEVRER and CLEVRER-Humans datasets

Enhancing the Security and Mitigating Bias in Vision Language Models (VLMs) to Combat Hateful Image Generation and Detoxify Hateful Images

Developing methods for preventing and mitigating toxic and biased content generation in VLMs through DE-HATE techniques.

QA-RAG: Leveraging Question and Answer-Based Retrieved Chunk Re-Formatting for Improving Response Quality During Retrieval-augmented Generation

Integrating QA-formatted databases to enhance context representation and response quality, supported by experimental evidence on benchmark RAG datasets, demonstrates improved performance compared to state-of-the-art text retrieval methods

Senior Engineer-Big Data & Data Science

SLT-MOBITEL - SRI LANKA

Involvement in Big Data & Data Science Project- Mobitel (Pvt) Ltd (Aug 2019 – July 2022)

Projects Highlights:

• Customer data churn prediction using deep learning.
• Broadband tower optimization using evolutionary algorithms.
• Customer segmentation using unsupervised machine learning.
• Seasonal recommendations & promotional campaigns using machine learning.
• Mobile fraud detection and prevention using deep learning.
• Alexa conversational AI chatbot using natural language processing and machine learning.
• mAdvertising platform using unsupervised learning.
• M2M IMSI classification using deep learning and unsupervised machine learning.
• Smartphone recommendation system using deep learning.
• Predictive analytics solution for international calling (IDD) & inbound roaming using machine learning.

Software Engineer-Artificial Intelligence

SimCentric Technologies - SRI LANKA

Projects Highlights:

SmartTrainer - An adaptive framework to prevent real time injuries.

• Smart wearable and Android based application for human activity recognition and alerting to prevent real time injuries.

A machine learning framework to train simulation AI using live data.

• DISAARPlugin Module to record DIS packets in simulation that is integrated with the SAF product at SimCentric Technologies.

Software Engineer Intern

WSO2

XACML Development Tool

Projects Highlights:

• XACML (eXtensible Access Control Markup Language) is a fine-grained access control language.
• It is not a simple task to write a policy from sketch without any support.
• Objective of this project was to develop a tool which compliance with XACML 3.0 specifications and introduce a UI based approach to create XACML policies and to evaluate them.

PhD in Computer Science

University of South Carolina

Grade Point Average: 3.88

MSC in Artificial Intelligence

University of Moratuwa - Sri Lanka

Grade Point Average: 3.6

BSC in Information Technology

General Sir John Kotelawala Defence University - Sri Lanka

Grade Point Average: 3.96 [FIRST CLASS & BATCH TOPPER]