Data Scientist / Data Engineer based in Skövde, Sweden. I turn raw, messy event data into production ML systems, and in my own time I build ML and analytics tools for the games I play.
Currently: MSc in Data Science at the University of Skövde, and looking for an MSc thesis in ML for games, especially mining gameplay telemetry to understand how players actually behave versus what designers intended.
My koi, redrawn nightly from my GitHub activity with git-koi.
| Project | What it is | Key result |
|---|---|---|
| osu-pulse | Real-time osu!mania telemetry (session > play > miss events) with an unsupervised ML layer: clustering, anomaly detection, regression | Every method checked against a baseline; random forest R² 0.71 vs 0.01 for the Poisson baseline |
| ManiaMapper | Generates playable osu!mania beatmaps from raw audio with a difficulty-conditioned encoder-decoder (BiLSTM + causal LSTM, ~7M parameters) | Shipped end to end as a FastAPI web app on GCP |
| mariokart8-build-prediction | Hand-built 52-track dataset predicting the best build per track | Found target leakage behind a fake R² of 0.99; honest result traced to players converging on 6 meta builds |
| k8s-traffic-autoscaling-lab | Local multi-node Kubernetes cluster with traffic-driven autoscaling (HPA) | Full setup and debugging docs |
Data Engineer at a healthtech company, remote, 2024 to present. HIPAA-compliant environment.
- Built an OCR/LLM extraction engine that cut manual data entry by about 88%
- Rebuilt the prediction pipeline: 15 minutes down to under 30 seconds per sample
- Behavioural analytics that flags at-risk clients from event patterns (17% of the client base)
- Took over infrastructure: CI/CD with preview environments cut deploys from a full day to about 30 minutes
- Honest evaluation first: baselines, cross-validation, and saying so when a model doesn't beat the baseline
- Fix the data at the source instead of working around it downstream
- Ship it: projects end as something people can run, not just a notebook
ML: PyTorch, scikit-learn, XGBoost, clustering, anomaly detection, sequence models (LSTM), feature engineering
Data: Python, SQL, Spark (Scala), pandas, ETL pipeline design, event and log data
Infra: Docker, Kubernetes, CI/CD (GitHub Actions), GCP, AWS, Redis
Rhythm games (osu!mania), Valorant, Hollow Knight and Silksong (modding), bouldering, basketball, strategy board games.
LinkedIn · aravinddora45@gmail.com · open to thesis and data/ML roles