Models on real data
Clinical measurement sets, image data and messy product listings — where the encoding and the class balance matter more than the algorithm.
Undergraduate · open to internships & collaboration
Thisara Anjana
- Machine Learning Engineer
- Data Science
- MLOps
Building prediction and detection models, data pipelines, and reproducible ML systems. I work in Python — training models on real datasets, then giving them the pieces they need to survive outside a notebook: versioned data, tracked experiments, a container, and somewhere to run.
Sri LankaPython · ML · MLOps
01About
I build machine learning systems, not just models.
ICT undergraduate at the University of Vavuniya, based in Sri Lanka, focused on machine learning, data engineering and MLOps.
Most of my early projects were notebooks — a dataset, a model, an accuracy number, and no way to get back to it a month later. The work since then has been about closing that gap: taking the same modelling and putting the surrounding system in place.
That meant learning the boring parts on purpose. Loading data as an Airflow DAG instead of a script. Declaring training stages in dvc.yaml so inputs are tracked. Logging runs to MLflow instead of comparing terminal output. Writing a Dockerfile so it runs somewhere other than my laptop.
Alongside that, my degree covers the general engineering base — databases, networking, systems and several languages beyond Python — which is why some of my repositories are a C# chat app or a Kotlin video player rather than a model.
Everything claimed on this page links to a public repository. If there is no repository behind it, it is marked as coursework instead.
Pipelines that re-run
Scheduled ETL into PostgreSQL, and training expressed as declared stages so a commit reproduces one exact result.
Models that leave the notebook
Tracked runs, versioned datasets, a Dockerfile, and a small app in front — so the work is usable by someone other than me.
02// Skills & Technologies
Technical Arsenal
Technologies, tools, and engineering concepts I use to build data-driven and machine learning systems.
28
Technologies
6
Domains
26
Backed by a repository
01ML / Data Science
02Data Engineering
03MLOps / Deployment
00
Everything
Every technology below is here because a repository, a certificate or my degree coursework stands behind it. No percentages, no ratings — click any card to see exactly where it was used.
07
Currently exploring
Kept deliberately separate from the list above — these are things I'm actively learning, not things I've shipped yet.
Advanced MLOps
Model registries, promotion between stages, and monitoring a deployed model rather than just shipping it.
Cloud ML deployment
Taking training and serving fully onto managed cloud services instead of a single AWS-backed tracking server.
Advanced Airflow workflows
Sensors, backfills, dynamic task mapping and error handling beyond a linear extract-transform-load DAG.
00Pipeline
How data becomes a deployed model.
The stack below is the one I actually run, in the order it runs. Select a stage to see what it does and which repository it comes from.
PostgreSQL · Data storage
Destination for the ETL pipeline — tables written and queried from Airflow tasks.
Airflow-ETL-Pipeline-with-Postgres →The lifecycle I follow, stage by stage
01
Data
Pull the raw dataset in as an ingestion step with a config file, not a hand-copied CSV.
- PostgreSQL
- Airflow
02
Exploration
Notebook first — distributions, missing values, whether the target is even learnable.
- Pandas
- Jupyter
03
Preprocessing
Cleaning and schema validation promoted out of the notebook into a src/ module.
- Pandas
- DVC
04
Feature engineering
Encoding and scaling saved alongside the model so inference uses the same transform.
- Scikit-learn
05
Model training
Hyperparameters live in params.yaml so a run is described by a file, not by memory.
- Scikit-learn
- XGBoost
06
Experiment tracking
Parameters, metrics and artifacts logged per run instead of compared by eye.
- MLflow
07
Versioning
Data and model files tracked as DVC pointers so a commit reproduces one exact state.
- DVC
- Git
08
Containerization
A Dockerfile pins the environment so it runs the same off my machine.
- Docker
09
Deployment
Served behind a small Flask app, with tracking and serving moved onto AWS.
- Flask
- AWS
03Selected work
Featured Projects
A selection of projects where I apply machine learning, data science, and software engineering concepts to practical problems.
- case study01
- Machine Learning
- Data Science
Multiple Disease Prediction System
Two separate clinical datasets, two separate models — and no way for anyone without a notebook to actually use either of them.
DatasetPreprocessingTrainingEvaluationSaved model + scalerWeb app
- Python
- Scikit-learn
- JavaScript
- HTML/CSS
- case study02
- Data Science
- MLOps
End-to-End Data Science Project
Research notebooks and shipped code usually diverge: the notebook explores, the app re-implements, and the two disagree.
Data ingestionData validationTransformationModel trainerEvaluationFlask app
- Python
- Flask
- Docker
- YAML
- Jupyter
- 03
- Machine Learning
- Data Science
Laptop Price Prediction
Laptop specification data is messy — memory, screen and CPU fields arrive as free text, not as numbers a regressor can use.
Raw listingsFeature engineeringEncodingRegressionEvaluation
- Python
- Pandas
- Scikit-learn
- Jupyter
- 04
- Machine Learning
- Data Science
Music Recommendation System
Recommendation has no single right answer, so it can't be evaluated the way a classifier is.
Track datasetFeature vectorsSimilarityRanked recommendations
- Python
- Pandas
- Scikit-learn
- Jupyter
04Credentials
Certifications and credentials, in the order I earned them.
Short courses, not degrees — every entry carries the actual certificate, its issuer, its date and its verification ID where one exists. Open one to read it here without leaving the page.
Credentials / index · 9 of 9
2026
2025
05Journey
Education, and the path I'm taking toward ML engineering.
No employment history here yet — I'm an undergraduate. What follows is the actual progression: each step names the work that carried it, with no invented dates or achievements.
Progression
01
ICT undergraduate
Started the ICT degree at the University of Vavuniya — programming, databases and networking as the base, with statistics and software design alongside.
Coursework — no repository attached.
02
Programming fundamentals
Small self-contained projects to get comfortable writing code that others run: a Sudoku solver, a Space Invaders clone, a desktop chat app and an Android video player.
03
Machine learning models
Supervised learning on real datasets — heart disease, Parkinson's and diabetes prediction, sonar rock-vs-mine classification, fake news detection and laptop price regression — plus a music recommender.
04
Data engineering
Moved from hand-run notebooks to scheduled work: an Airflow DAG doing extract, transform and load into PostgreSQL, running under docker-compose.
05
Reproducible pipelines
Staged training with DVC so preprocess → train → evaluate re-runs from a commit, with data and models tracked as pointers rather than blobs.
06
MLOps & deployment
Experiments tracked in MLflow, the app packaged in Docker behind Flask, tests wired into GitHub Actions, and tracking plus serving moved onto AWS.
Education
BSc (Hons) Information and Communication Technology
University of Vavuniya, Sri Lanka
2023 – 2027 (expected)
Undergraduate degree covering programming, databases, networking and systems, alongside self-directed work in machine learning and MLOps.
Relevant areas
- Programming
- Databases
- Networking
- Software design
- Statistics
- Systems
06Contact
Open to internships, collaboration and interesting datasets.
Whether it's an ML internship, a data pipeline that needs building, or a project you want a second pair of hands on — send it over.