I learn things by taking them apart.
I'm Kushan Manahara, a machine learning engineer. I write here because the notes are more useful in public than sitting in a private file.
What I write about
Four things, roughly.
Machine learning and AI, especially what is actually happening underneath the abstractions — from mathematical foundations and model training to neural networks, inference, fine-tuning, and modern LLM systems.
AI agents and the systems around them — tool calling, MCP, RAG, agent architectures, orchestration, evaluation, and the protocols that connect models to real software, data, and services.
The engineering underneath AI — Python, TypeScript, Linux, distributed systems, cloud infrastructure, containers, Kubernetes, Kafka, APIs, databases, and the tooling needed to turn an AI prototype into something that can actually run in production.
Milestones and lessons from the journey — things I’ve learned while moving from software engineering and AI/ML research into machine learning engineering and production AI systems.
Most of what I write starts as something I’m trying to understand myself. I learn by taking things apart, following the reasoning all the way down, building something with it, and writing down what I wish I had understood the first time.
Milestones
Where the engineering side of this actually got its footing.
2026
Machine Learning Engineer
H2O.ai
Working on production AI and ML systems across LLM applications, agentic systems, predictive AI, AutoML, and the H2O ecosystem.
2026
Machine Learning Engineer
CML Insight
Worked on machine learning systems and applied ML engineering, bridging research ideas with practical production-oriented solutions.
2025
AI/ML Research Assistant
University of Peradeniya
Worked on AI/ML research while exploring problems at the intersection of machine learning, software engineering, and research.
2024
B.Sc. (Hons) in Engineering
University of Peradeniya
Graduated in Computer Engineering, building a foundation across software engineering, systems, mathematics, and artificial intelligence.
2024
Full-Stack Software Engineer Intern
GTN Technologies
Worked on full-stack software engineering in a production fintech environment, gaining practical experience across frontend, backend, APIs, databases, and application development.
2024
Final Year Project
University of Peradeniya
Developed an Oxford Nanopore Technology-based pipeline for RNA-Seq data analysis, working with Chiran Govinna and Tharindu Dhananjaya in a research area that was new to the team.
The setup
- OS
- Linux & macOS
- Distros
- Ubuntu, Debian, Kali
- Languages
- Python, TypeScript, JavaScript, Java, Go, SQL
- ML
- PyTorch, TensorFlow, scikit-learn
- AI
- LLMs, RAG, AI Agents, MCP, LangGraph
- Infrastructure
- Docker, Kubernetes, Kafka, AWS & Google Cloud
How I publish
One post when it’s ready, never on a schedule.
Most posts begin as something I’m learning, debugging, building, or trying to explain to myself. I prefer understanding the mechanism over memorizing the terminology.
When a post relies on someone else’s research, numbers, or findings, I try to make the source clear.
Say hello
Corrections, questions, ideas, and disagreements are all welcome — especially the last one.