Supul Pushpakumara | AI Engineer & ML-Ops Architect

About

Who I Am

Recent Work

Engineering Philosophy

"Ship fast, but ship right. Production-grade means thinking about scale, security, and maintainability from day one."

AI Engineer & ML-Ops Specialist

I'm an AI Engineer who bridges the gap between research and production. While many can build models, I specialize in making them reliable, scalable, and production-ready. My journey has taken me from academic research in document understanding to architecting enterprise-grade AI infrastructure at IFS R&D International.

What drives me is the challenge of taking cutting-edge AI research and transforming it into systems that businesses can depend on 24/7. This means obsessing over details that many overlook: container optimization, CI/CD pipeline efficiency, service authentication, and cost management. I believe that great AI infrastructure is invisible—it just works, scales automatically, and doesn't keep engineers awake at night.

The best ML-Ops work is the kind that nobody notices because everything runs smoothly. My goal is to build infrastructure that empowers data scientists to focus on what they do best—building models—while the deployment, scaling, and monitoring happen seamlessly in the background.

Company: IFS R&D International
Degree: BSc Engineering (Hons)
Location: Sri Lanka

CGPA: 3.85 / 4.2 (First Class)
University: University of Moratuwa
Stream: Data Science Engineering

What I Value Most

Performance at Scale

Security-First Design

Continuous Improvement

Knowledge Sharing

Get In Touch LinkedIn Profile Download CV

92%

Deployment Time Reduction

10x

Code Coverage Improvement

A

SonarCloud Rating

1.5+

Years at IFS

Experience

Professional Journey

IFS R&D International

Data Scientist June 2024 - Present

At IFS, I've taken on a role that goes far beyond traditional data science. While my title says "Data Scientist," my day-to-day work involves architecting and building the infrastructure that powers AI services across the organization. I joined during a critical phase of growth where the AI team needed someone who could bridge the gap between research prototypes and production-ready systems.

One of my first major initiatives was tackling the developer environment problem. Previously, each developer had their own Kubernetes namespace with duplicated infrastructure—inefficient, costly, and a maintenance nightmare. I designed and implemented a shared infrastructure model that reduced deployment time from nearly an hour to just 5 minutes, a 92% improvement that fundamentally changed how the team works.

Beyond infrastructure, I've been the driving force behind introducing modern engineering practices to the team. When I arrived, code coverage was at a concerning 8%. Through systematic implementation of testing frameworks, code review processes, and quality gates, I helped elevate this to 80%—earning us 'A' ratings across SonarCloud's security and maintainability metrics. This wasn't just about numbers; it was about building a culture of quality that will pay dividends for years.

Key Highlights - What Sets Me Apart

Cost Optimization Leader

Architected shared infrastructure replacing the per-developer namespace approach. This wasn't just about cost savings—it was about creating a consistent, reproducible environment that eliminated "works on my machine" issues. The 92% reduction in deployment time (from 1 hour to 5 minutes) transformed developer productivity across the team.

Tool Introduction Pioneer

Successfully championed the adoption of SonarCloud, Veracode, KServe, Tekton, and UV package manager. Each tool required building consensus, creating documentation, and running training sessions. The UV package manager alone reduced Python dependency resolution from minutes to seconds, a game-changer for our CI pipelines.

Code Quality Champion

Transformed the team's approach to quality by implementing comprehensive testing strategies. We went from 8% to 80% code coverage—a 10x improvement. But more importantly, I established a culture where quality is everyone's responsibility, with automated gates that catch issues before they reach production.

Full-Stack ML-Ops Architect

Designed production-grade microservices with dual protocol support (HTTP/gRPC), implementing service authentication using Kubernetes ServiceAccount tokens and multi-region load balancing. This architecture handles millions of inference requests while maintaining sub-100ms latency at the 99th percentile.

Technical Deep Dives

Repository Migration & DevOps Transformation

Led a major repository migration from legacy systems to a modern Git-based workflow with zero service disruption. This wasn't a simple copy operation—it required designing automated synchronization pipelines that maintained history, handled branch protection rules, and integrated with our CI/CD systems. The migration enabled modern practices like pull request workflows, automated code review, and seamless deployment triggers.

Zero-Trust Service Authentication

Implemented service-to-service authentication using Kubernetes ServiceAccount tokens, eliminating the need for static credentials. Each service automatically receives short-lived tokens that are validated by the receiving service, implementing a zero-trust architecture where every request is authenticated. This approach significantly reduced our attack surface and simplified credential management.

Go-Based Microservice Architecture

Designed and built a Go-based microservice architecture supporting both HTTP and gRPC protocols. This dual-protocol approach allows internal services to communicate via efficient gRPC while exposing REST APIs for external consumers. The architecture includes intelligent routing for multiple LLM providers, automatic failover, and request-level metrics collection for observability.

Technologies in Daily Use

Go Python Docker Kubernetes Tekton Azure KServe Veracode

CML Insight

ML and Software Engineer Jan 2023 - Oct 2023

My time at CML Insight was foundational in shaping my understanding of production ML systems. Working on Google Cloud Platform, I gained hands-on experience with cloud-native data processing using Dataflow and Workflow pipelines. This exposure to enterprise-grade data infrastructure taught me the importance of thinking about scale and reliability from the start.

I also contributed to frontend development using React.js and Material-UI, which gave me appreciation for the full stack. Understanding how data scientists and analysts interact with ML systems through UIs has made me a better infrastructure engineer—I now build systems with the end user experience in mind.

Part Time Engineer July - October 2023

Continued contributing while pursuing my final year of studies. Focused on Google Cloud Platform architecture, building robust dataflow and workflow pipelines that processed terabytes of data daily. Also contributed to the React.js frontend, implementing data visualization components using MUI.

ML Engineering Intern January - June 2023

Developed Python data flow pipelines using Apache Airflow and Dagster—two of the most popular workflow orchestration tools in the ML ecosystem. This experience taught me the importance of reproducibility, idempotency, and proper error handling in data pipelines. I also gained hands-on experience with various ML libraries and developed an appreciation for clean, well-documented code.

Tech Stack

Technologies & Tools

My toolkit has evolved through real-world problem solving. Rather than collecting certifications for their own sake, I've focused on mastering technologies that solve actual production challenges. Here's what I work with daily and why each tool matters in the modern AI infrastructure stack.

Languages

Go for high-performance services, Python for ML, and a range of languages for full-stack development.

Go Python C# JavaScript Java SQL PHP HTML/CSS

Cloud & Kubernetes

Building scalable infrastructure on Azure and GCP with Kubernetes as the orchestration layer.

AKS Helm GCS KEDA Azure GCP ACR Key Vault Heroku

CI/CD & MLOps

Automating everything from code commit to production deployment with GitOps principles.

Tekton ArgoCD Apache Airflow Dagster Bitbucket Pipelines KServe Kubeflow Terraform

Containerization

Crafting minimal, secure container images using multi-stage builds and distroless base images.

Docker k9s Multi-stage Builds Distroless Images

Security & Quality

Security isn't an afterthought—it's baked into every pipeline and service from day one.

SonarCloud Veracode Keycloak OAuth2 JWT RBAC

APIs & Protocols

Building efficient service communication with gRPC for internal and REST for external APIs.

gRPC REST HTTP/HTTPS Service Mesh XML

Observability

You can't improve what you can't measure. Full visibility into every service and request.

Prometheus Grafana Logging Tracing

ML & AI

From model training to production serving, covering the full ML lifecycle.

PyTorch Transformers TensorFlow Scikit-learn Pandas NumPy LLM Providers Model Serving

Frontend Frameworks

Building responsive user interfaces and interactive dashboards for data visualization.

React.js MUI Streamlit jQuery Bootstrap

Backend Frameworks

Server-side development with modern frameworks for APIs and web applications.

FastAPI Flask Node.js Express Go (net/http)

Databases

Working with both SQL and NoSQL databases for various application needs.

PostgreSQL MySQL MongoDB Firebase Firestore

Data Visualization

Creating insightful visualizations and interactive dashboards for data-driven decisions.

PowerBI Plotly Tableau Streamlit ArcGIS

Dev Tools & Collaboration

Version control, project management, and collaboration tools for team productivity.

Git GitHub BitBucket Jira Confluence ClickUp Android Studio

Research & Projects

Notable Work

Beyond my professional work, I'm passionate about research and building projects that push boundaries. My research focuses on bringing AI capabilities to underserved languages, while my side projects explore practical applications of emerging technologies.

Research Paper

SinFUND & SinOCR

Benchmarks for Sinhala Handwritten OCR and Template-Free Form Understanding

This research addresses a critical gap in AI accessibility: the lack of quality datasets for low-resource languages. Sinhala, spoken by over 20 million people, had virtually no public benchmarks for document understanding tasks. We created two datasets that will enable researchers worldwide to develop and evaluate OCR and form understanding systems for Sinhala.

SinFUND is the first fully annotated dataset of 100 diverse Sinhala forms with bounding boxes, text transcriptions, and semantic labels. SinOCR contains 100,000 images covering handwritten texts across 200 different Sinhala fonts, enabling robust training of recognition models.

OCR Token Classification Document Understanding PyTorch Transformers

Supervisor: Dr. Thanuja Ambegoda July 2023 - Jan 2024

View Research Paper

Safe TakeOff

Weather Forecasting and Aircraft Decision Making

Built a comprehensive web application that helps aviation professionals make informed takeoff and landing decisions. The system uses time series analysis to forecast weather conditions and presents insights through an intuitive dashboard. The project demonstrated how ML can be applied to safety-critical decision making.

React Flask TensorFlow PowerBI MongoDB

Customer Churn Prediction

Interactive Dashboard for Business Analytics

Developed an end-to-end solution for predicting customer churn, from data preprocessing to a polished Streamlit dashboard. The project compared multiple ML algorithms and provided business-friendly visualizations that helped non-technical stakeholders understand and act on predictions.

Python Streamlit Plotly Pandas

ChallengerX Incubator

Disaster Risk Analysis & Asset Management

An incubator project developed in collaboration with IFS and Hatch, focused on using ML to improve disaster response. The system analyzes risk factors, predicts resource needs, and helps coordinate asset distribution using geospatial analysis through ArcGIS integration.

React ArcGIS FastAPI Firebase ML

HR Management System

Full-Stack Employee Management Solution

Complete HR management system with database backend. Features include employee leave applications with supervisor approval workflows, HR manager controls, and admin dashboards with report generation. Implemented SQL triggers and procedures with Azure deployment.

React Node.js Express MySQL Azure

Supplier Management System

Enterprise Supplier Database Application

Web application for managing supplier relationships with full CRUD operations. Supply managers can add, view, edit, and delete suppliers. Role-based access control for administrators and managers. Deployed on Heroku.

React Node.js Express MySQL Heroku

E-Pharmacy

Online Pharmacy Platform

Online pharmacy application where customers can upload prescriptions and order medicine. Admin dashboard for order management and prescription-based medicine dispensing. Built with Firebase backend.

React Firebase Firestore JavaScript

E-nstitute

Tutor Discovery Platform

Platform connecting students with private tutors across Sri Lanka. Teachers and students register and search based on location and subject. Features rating and review system, group management, and location-based search. Built from scratch without frameworks.

PHP HTML/CSS JavaScript MySQL

Education

Academic Background

BSc Engineering (Hons)

Computer Science & Engineering

University of Moratuwa 2019 - 2024

Graduated with First Class Honours from Sri Lanka's premier engineering university. My specialization in Data Science Engineering provided a strong foundation in both theoretical computer science and practical machine learning. The rigorous curriculum, combined with industry internships, prepared me for the challenges of building production AI systems.

CGPA: 3.85 / 4.2 (First Class Honours)
Stream: Data Science Engineering

GCE Advanced Level

Physical Science

Royal College, Colombo 07 2015 - 2018

Mathematics: A | Chemistry: A | Physics: B

Certifications

Professional Development

I believe in continuous learning. These certifications represent structured learning paths that complemented my hands-on experience, providing theoretical foundations for the technologies I use daily.

Machine Learning Specialization Coursera Deep Learning Specialization Coursera Google Hash Code 2021 Google SQL Coursera Git and GitHub Google | Coursera C Programming Coursera JavaScript, jQuery, JSON Coursera

Leadership & Community

Rotaract Club of Alumni - UoM

Sergeant at Arms (2023)

Responsible for maintaining order, security, and protocol within the club.

Rotaract Club of UoM

Senior Director (2022-2023)

Advisor to the director board and executive committee member.

Environmental Services

Director (2021-2022)

Led award-winning projects: Zooxanthellae (Coral Restoration) & Grove Green (Mangrove Restoration).

OREPA

Member, Student Chapter (2020-Present)

Old Royalists Engineering Professionals Association.

Contact

Get In Touch

I'm always interested in discussing AI infrastructure, ML-Ops best practices, or potential collaborations. Whether you have a challenging technical problem or just want to chat about the future of AI systems, feel free to reach out.

Email

supul9229kalhara@gmail.com

Phone

+94 71 518 0267

LinkedIn

Connect with me

GitHub

View my code

References

These professionals have worked closely with me and can speak to my technical abilities, work ethic, and collaborative approach.

Jayson Castro

Senior Lead Machine Learning Engineer

IFS R&D International

jayson.castro@ifs.com

Crishmi Costa

Senior Manager, AI Program Delivery

IFS AI - BU Investment

crishmi.costa@ifs.com

Dr. Thanuja Ambegoda

Senior Lecturer, CSE

University of Moratuwa

thanuja@cse.mrt.ac.lk

Supul's AI Assistant

Online

Hi! 👋 I'm Supul's AI assistant. I can answer questions about his experience, skills, projects, and more. What would you like to know?