About
Hi, I'm Kushan. I build intelligent systems.
AI/ML engineer in Sri Lanka. Machine Learning Engineer at H2O.ai.
My work spans research and production, from experimenting with ML models and LLM-based systems to designing APIs, data pipelines, agent workflows and scalable cloud infrastructure.
I care deeply about understanding how systems work under the hood. Rather than treating AI as a black box, I like breaking problems down, understanding the trade-offs and turning ideas into reliable software.
My experience has taken me from full-stack development and university research into machine learning engineering and forward-deployed AI systems.
- Based in
- Sri Lanka, building toward global AI/ML engineering work.
- Currently
- Machine Learning Engineer at H2O.ai, working on forward-deployed AI engineering.
- Focus
- Intelligent production systems combining AI/ML, LLMs, agents, backend infrastructure and cloud.
- Education
- B.Sc. Engineering (Hons) in Computer Engineering, University of Peradeniya.
What I work on
- 01AI / Machine LearningBuilding production ML systems from data and experimentation through model serving, evaluation and deployment.Machine LearningDeep LearningPyTorch
- 02LLMs & AI AgentsDesigning intelligent applications around LLMs, RAG, tool use, agent workflows and multi-step reasoning.LLMsRAGAI Agents
- 03Backend & Distributed SystemsEngineering reliable backend systems, APIs, data pipelines and distributed workflows that support production AI applications.PythonNode.jsKafka
- 04Cloud & MLOpsTaking AI systems from local experiments to secure, scalable cloud infrastructure and production workflows.AWSGCPMLOps
- 05Full-Stack EngineeringBuilding complete products across frontend, backend, databases and intelligent services rather than treating AI as an isolated component.ReactNext.jsTypeScript
How I think
Five rules I work by.
- 01Understand the systemI want to understand what happens underneath the abstraction, not simply learn how to call an API.Digging into tensor memory layout, strides and training mechanics instead of stopping at the framework API.
- 02Build end to endA model is only one part of an AI product. I think about the complete path from user interaction and APIs through data, models, infrastructure and production behaviour.My AI projects span RAG and LangGraph workflows through databases, authentication, live event streams and the frontend.
- 03Reason before choosingI prefer understanding why a technology or algorithm is appropriate instead of memorising that it is the conventional choice.Learning algorithms by their behaviour, complexity and trade-offs before their implementation patterns.
- 04Production changes the problemA prototype can demonstrate that something works. Production engineering asks whether it stays reliable when users, data, failures and edge cases appear.Chasing down duplicate detections, image orientation bugs, import/export mismatches and missing tests, not just the model.
- 05Learn by buildingThe fastest way I internalise a concept is to use it to build something real.Full-stack apps, then research systems, then ML engineering, LLM applications and AI agents.
’202020 — 2024 · PeradeniyaUniversityB.Sc. Engineering (Hons), Computer Engineering, University of Peradeniya’232023 — 2024 · Sri LankaFull-Stack EngineeringFull-stack development intern, GTN Technologies’242024 — 2025 · PeradeniyaResearchAI/ML research assistant, University of Peradeniya’242024 — 2026 · RemoteMachine Learning EngineeringMachine learning engineer, CML Insight’262026 — now · Sri LankaAI EngineeringMachine learning engineer, H2O.ai
Let's build