Auckland, New Zealand

Malitha Gunawardhana

Machine Learning Engineer | Data Scientist | Applied AI Researcher

PhD-trained AI and machine learning professional with 5+ years of experience spanning production machine learning, deep learning, computer vision, data science, generative AI, time-series analytics and software engineering. I build end-to-end systems from experimentation and statistical evaluation through deployment, monitoring and real-world use.

My work combines research-grade experimentation with software engineering and MLOps across Python, PyTorch, SQL, FastAPI, Docker, AWS, MLflow, Airflow, Kafka and modern foundation-model and agentic AI tooling.

AI & Machine Learning Data Science RAG & Agents Production ML Computer Vision

Email Download CV LinkedIn GitHub Google Scholar

Open to opportunities Machine Learning · Data Science & Analytics · Applied AI · Research

5+ YearsAI, ML & Software

PhDDeep Learning & Computer Vision

Machine LearningResearch to production

Data ScienceStatistics, experimentation & analytics

Applied AIComputer vision, GenAI & sensor systems

Career

Experience

Industry and research experience across production machine learning, medical AI, computer vision, sensor analytics and software engineering.

Machine Learning Engineer (Part-Time)

Viotel Pty Ltd · Auckland, New Zealand

Mar 2026 - Sep 2026

Developed and deployed an end-to-end deep-learning system for automated seismic phase identification across 3,700+ events and 30,000+ station samples, reducing geophysicists' manual processing effort by more than 70%.
Analysed high-frequency accelerometer data across 10+ telecommunications towers, combining weather and environmental context with controlled experiments to distinguish abnormal vibration and climbing activity from normal operational variation.
Designed the seismic data and modelling workflow across 17 geothermal sensors, covering dataset and label preparation, model comparison, expert-pick validation, quality checks and scheduled inference on AWS-hosted data.
Identified a data-duplication defect in the Protobuf-based accelerometer conversion pipeline, traced the issue to conversion logic and worked with the hardware team to support resolution.
Packaged recurring inference with Docker and built a dashboard comparing automated and expert phase picks, confidence and timing errors to support prediction-quality review and threshold refinement.

Deep LearningTime SeriesSignal ProcessingAWSDocker

IPPT-PAN

Institute of Fundamental Technological Research, Polish Academy of Sciences · Warsaw, Poland

Sep 2022 - Jan 2025

Machine Learning Engineer

Nov 2023 - Jan 2025

Developed and refined deep-learning models for breast-tumour segmentation, benign/malignant classification and BI-RADS category prediction using a multi-centre dataset of 3,000+ ultrasound images, achieving 86%+ Dice for tumour segmentation.
Benchmarked modern medical-imaging approaches and adapted a pretrained foundation segmentation model using parameter-efficient fine-tuning for ultrasound lesion segmentation.
Designed reusable machine-learning pipelines covering preprocessing, augmentation, training, validation, inference and evaluation using PyTorch, AWS and Docker.
Diagnosed shortcut learning and model risk linked to scanner markers, biopsy needles and other imaging artefacts, guiding data and model improvements for more reliable cross-site behaviour.

PyTorchMedical ImagingFoundation ModelsAWSDockerML Pipelines

Data Scientist

Sep 2022 - Nov 2023

Contributed to the INFOSTRATEG research project developing AI-assisted methods for breast-cancer lesion analysis using medical ultrasonography.
Curated and analysed multi-centre ultrasound datasets, covering data sourcing, preprocessing, augmentation, data-quality checks, class-imbalance handling and patient-level dataset preparation.
Developed and compared deep-learning approaches for breast-tumour detection, segmentation and classification, testing architectures, loss functions, augmentation strategies and hyperparameters.
Evaluated models using cross-validation, sensitivity/specificity analysis and clinically focused false-positive/false-negative review.
Worked with medical researchers and clinical collaborators to translate clinical requirements into measurable machine-learning objectives, evaluation criteria and experimental designs.

Data ScienceDeep LearningComputer VisionClinical AIModel Evaluation

Artificial Intelligence Researcher

Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) · Abu Dhabi, UAE

Sep 2022 - Sep 2023

Co-developed a train-time object-detection calibration method addressing class confidence and bounding-box localisation, contributing to a CVPR 2023 publication.
Contributed to semi-supervised domain-generalisation research combining limited labelled data with unlabelled samples, delivering improvements across five benchmarks and four semi-supervised baselines and contributing to a CVPR 2024 publication.
Developed and evaluated self-supervised video representations, studying temporal representation learning and transfer to downstream video-understanding tasks under limited annotation.

CalibrationSemi-Supervised LearningSelf-Supervised LearningComputer VisionResearch

Machine Learning Engineer

PromiseQ GmbH · Berlin, Germany

Jun 2022 - Nov 2022

Developed and optimised real-time computer-vision models for an AI-surveillance platform operating across 3,000+ CCTV cameras, covering people and construction-vehicle detection, intrusion and other safety/security use cases.
Reduced false alarms by more than 5% through model calibration, retraining and improvements to detection, tracking and post-processing logic.
Analysed recurring production failures under poor lighting, occlusion, unusual poses, camera positioning, crowded scenes and motion blur, translating findings into operational thresholds and deployment actions.
Monitored deployed-model performance, input quality, drift, confidence, latency, incidents and model versions across GCP, AWS, streaming data, APIs and CI/CD workflows.

Computer VisionObject DetectionModel MonitoringGCP / AWSCI/CD

Full-Stack Software Engineer

Xeptagon (Pvt) Ltd · Colombo, Sri Lanka

Mar 2021 - May 2022

Engineered a low-latency domain drop-catching platform using Go, MongoDB, REST APIs, TypeScript and JavaScript, achieving a 90%+ successful domain catch rate with successful catches completed within approximately 50 ms of availability.
Applied statistical and historical analysis to identify high-value domain opportunities, timing patterns and competition behaviour, linking analytical findings to commercially valuable decisions.
Optimised command-sending logic by balancing timing, command/rate constraints, execution cost and probability of success, informing infrastructure and operating-strategy decisions.
Translated requirements from CEOs, CTOs and managers into APIs, database schemas, logging and production tooling while directing a four-member engineering team across architecture, planning, documentation and code review.

GoMongoDBREST APIsLow-Latency SystemsData Analysis

Builds

Featured Projects

Selected projects that demonstrate modern AI application engineering, data science, RAG, agents and production MLOps. Full implementations are available through GitHub.

Agentic AI · RAG · Voice · Cloud

Healthcare Voice & Chat Assistant

Built an end-to-end text and voice healthcare assistant that answers hospital-information queries and supports appointment/CRM workflows. The system combines LangGraph agents, MCP tools, Qdrant/pgvector RAG, multi-layer memory, Redis background processing, streaming voice with LiveKit/Deepgram/ElevenLabs, Langfuse observability and AWS deployment.

LangGraphMCPQdrantpgvectorLangfuseAWS ECS/Fargate

Data Science · Data Engineering · MLOps

ChurnOps

Built an end-to-end customer churn prediction and MLOps system spanning data validation, preprocessing, model selection and evaluation through batch and Kafka-based streaming inference. Added PySpark pipelines, MLflow tracking and model registration, Airflow orchestration, drift monitoring, PostgreSQL/RDS analytics writes and Dockerised AWS deployment.

PySparkKafkaMLflowAirflowDockerAWS

Document AI · Human-in-the-Loop · Backend Systems

ClaimLoop

Built a production-ready human-in-the-loop insurance claims workflow that extracts structured fields from PDFs with vision-capable LLMs, assigns confidence scores and routes uncertain cases to reviewers. Added PostgreSQL-backed task coordination, correction history, FastAPI and React services, concurrency safeguards, automated tests and Docker-based deployment.

Document AIFastAPIPostgreSQLReactDockerTesting

Foundation Models · Fine-Tuning · Multimodal · GraphRAG

Foundation Model Fine-Tuning & Advanced GenAI

Fine-tuned an open-source instruction-tuned language model using parameter-efficient 4-bit supervised fine-tuning, evaluated it with automated metrics, LLM-as-a-Judge and safety red-team testing, and built multimodal text-to-image retrieval and a knowledge-graph RAG prototype for multi-hop reasoning.

Fine-TuningQuantisationHugging FaceMultimodal RetrievalNeo4jGraphRAG

View all projects on GitHub

Academic foundation

Education

Research training in deep learning, computer vision, data science, uncertainty, robustness and medical AI, supported by an engineering foundation.

Doctor of Philosophy

University of Auckland · Deep Learning, Computer Vision & Data Science

Dec 2023 - Aug 2026

Thesis: Robust, Uncertainty-Aware, and Scalable Deep Learning for Bi-Atrial Segmentation from LGE-MRI in Atrial Fibrillation.

Developed a deep-learning framework for bi-atrial segmentation from 3D cardiac MRI, addressing cross-hospital generalisation, severe class imbalance and thin anatomical boundaries.

Evaluated robustness, uncertainty and model risk under distribution shift using controlled experiments, confidence intervals, statistical testing and clinically meaningful failure analysis across multi-centre cohorts.

Developed a 3D self-supervised learning framework combining masked autoencoding and contrastive representation learning to reduce dependence on expert annotations, supported by HPC, multi-GPU training and reproducible workflows.

Award: Health Research Council Scholarship.

Research publications across my PhD and earlier research roles: peer-reviewed work in CVPR, Medical Image Analysis, MIDL and other venues. View all publications on Google Scholar →

B.Sc. Engineering Honours

University of Moratuwa · Sri Lanka

Jan 2017 - Jul 2021

Engineering foundation spanning mathematics, machine learning, image processing and machine vision, signals and systems, algorithms, operations research and project management. Dean's List, Semester 7.

Technical toolkit

Skills

Focused on the stack I use to move from data and experimentation through reliable AI systems, analytics and production deployment.

AI & Machine Learning

Python, PyTorch, TensorFlow, scikit-learn, XGBoost, LightGBM, deep learning, computer vision, object detection, medical imaging, time-series modelling, anomaly detection, self-supervised learning and semi-supervised learning.

Foundation Models & Generative AI

Hugging Face Transformers, PEFT, LoRA, supervised fine-tuning, quantisation, prompt engineering, multimodal retrieval, structured outputs and LLM evaluation.

RAG, Agents & LLMOps

RAG, BM25, embeddings, hybrid retrieval, reciprocal-rank fusion, cross-encoder reranking, LangChain, LangGraph, CrewAI, MCP, GraphRAG, Neo4j/Cypher, Qdrant, pgvector, Langfuse, LiveKit, Deepgram and ElevenLabs.

Data Science & Analytics

Python, SQL, Pandas, PySpark, statistical modelling, hypothesis testing, experimental design, confidence intervals, feature engineering, cross-validation, Power BI, Tableau and Excel.

Reliable & Responsible AI

Uncertainty quantification, calibration, robustness, generalisation, causal reasoning, error analysis, bias analysis, model risk, data governance, privacy and ethics-aware evaluation.

Production ML & MLOps

MLflow, Airflow, Kafka, Docker, GitHub Actions, experiment tracking, model registry and versioning, automated training and inference, data-quality checks, drift monitoring and observability.

Cloud & Software Engineering

AWS, GCP, Azure, FastAPI, REST APIs, PostgreSQL/RDS, Redis, MongoDB, React, TypeScript, JavaScript, Go, Linux, CI/CD and low-latency backend systems.

Professional development

Certifications & Training

Selected technical and professional training supporting my work across AI, cloud, data science and applied engineering.

Oracle

Oracle Cloud Infrastructure Generative AI Certified Professional

Training and certification covering large language model fundamentals, fine-tuning, generative AI services, deployment and production-oriented GenAI workflows.

UC Berkeley

Large Language Model Agents

Advanced training in LLM agents, multi-agent systems, retrieval-augmented generation, tool-enabled workflows, model deployment and MLOps for agentic applications.

Microsoft

Microsoft Certified: Azure Fundamentals (AZ-900)

Foundation-level certification covering Azure cloud services, compute, storage, networking, security, governance and core cloud concepts.

DeepLearning.AI

Deep Learning Specialization

Training across neural networks, optimisation, convolutional neural networks, recurrent neural networks, sequence modelling and practical deep-learning development.

McKinsey & Company

McKinsey.org Forward Program

Professional-development programme covering structured problem solving, communicating for impact, adaptability and resilience, digital and AI essentials and self-leadership.

365 Data Science

Data Science Career Track

Applied training covering Python, SQL, data wrangling, statistical analysis, machine learning, model evaluation and data-driven analytical workflows.

Hato Hone St John

First Aid Level 1

NZQA Unit Standards 6401 and 6402 covering first aid, CPR and basic life support, AED use, choking, bleeding, shock and common medical emergencies.

Beyond engineering

Service & Leadership

Selected leadership and mentoring experience alongside technical work and research.

2024 - 2026

Student Council Member & Postgraduate Committee Representative

Auckland Bioengineering Institute, University of Auckland

Represented postgraduate students in discussions with academic staff and institute leadership, raising student concerns and contributing to postgraduate policy, welfare and institute-level initiatives.

2024 - 2025

Mentor

Sustainable Education Foundation · ScholarX

Mentored 8+ university students on final-year research, higher-study opportunities, career planning and transitions between academia and industry through structured one-to-one guidance.

2021 - 2024

Vice President, Club Service

Rotaract Club of Alumni of University of Moratuwa

Progressed from Co-Director to Vice President and served two consecutive VP terms, leading Club Service within a 100+ member organisation, overseeing 15+ initiatives and 29+ project events/phases, coordinating with 12+ Rotaract clubs and receiving Spirit of Service recognition; the Let's Meet initiative received a district-level Silver Award.