Piotr Parkitny
Senior Director, Data & Analytics. Analytics leadership, data platforms, experimentation, forecasting, and ML at scale — from research to production.
Piotr Parkitny
Senior Director, Data & Analytics · Toronto, Canada
Turning raw data into decisions: analytics leadership, data platforms, experimentation, forecasting, and ML at scale. 15+ years shipping data & ML systems from research to production. MIDS at UC Berkeley.
Leadership & Core Skills: Analytics Strategy · Experimentation (A/B Testing) · Forecasting · Statistical Inference · Data Platforms & Pipelines · Spark · AWS · Python · AI/ML Systems · LLMOps · Edge AI
LinkedIn · Email · GitHub · Kaggle
Featured Engineering Projects
01 · Distributed Data Platform — Reproducible, scalable market-data platform: ingests raw stock-minute data from object storage, validates and transforms it into analytical datasets, and serves them through distributed and local analytical engines. code
02 · VocalForge — Tencent AuK 1.5B speech generation & editing (TTS, content/acoustic/paralinguistic editing, enhancement, separation) on Kaggle’s free 2× T4 GPUs, with weights served from Kaggle datasets. code · Kaggle
03 · EdgeGuardAI — Modular, event-driven security system combining local face recognition and multi-sensor correlation to publish automated voice announcements via MQTT. code
Earlier AI / ML Work
- 04 · Face Emotion Detection → Edge — real-time emotion recognition on an Nvidia Jetson: AWS training → MQTT → BigQuery dashboard. code · Kaggle
- 05 · MRI Analyzer — AI-assisted prostate cancer diagnosis: DICOM extraction, DNN ensemble with uncertainty. code · Kaggle
- 06 · Centralized Medical Data Platform — petabyte-scale, HIPAA-compliant health-record lake: Kafka → Spark Streaming → Parquet/ORC → Presto. code · slides
- Attractiveness Bias — A/B Test at Scale — randomized controlled trial on LinkedIn profile photos; logistic regression with clustered standard errors. code · report · Kaggle
- Counterfeit — Sentiment-Enhanced Summarization — sentiment signals in attention. paper · code
- YOLO Object Detection — real-time multi-object detection in video. Kaggle
- Human Pose Estimation — MediaPipe pipeline with dual-view comparison. code
Research & Experiments
Research notebooks
- [Conversion A/B Curves](https://github.com/pparkitn/kaggle/blob/main/conversion-curve-a-b-test.ipynb) · [Kaggle](https://www.kaggle.com/code/dsptlp/conversion-curve-a-b-test) - [Prophet — Canadian CPI](https://github.com/pparkitn/kaggle/blob/main/timeseries-forecast-using-prophet.ipynb) · [Kaggle](https://www.kaggle.com/code/dsptlp/timeseries-forecast-using-prophet) - [Clustering — KMeans + PCA](https://github.com/pparkitn/kaggle/blob/main/clustering-kmeans-with-pca.ipynb) · [Kaggle](https://www.kaggle.com/code/dsptlp/clustering-kmeans-with-pca) - [Classification → SQL](https://github.com/pparkitn/kaggle/blob/main/classification.ipynb) · [Kaggle](https://www.kaggle.com/dsptlp/classificationtosql) - [LDA vs PCA](https://github.com/pparkitn/kaggle/blob/main/lda-vs-pca.ipynb) · [Kaggle](https://www.kaggle.com/code/dsptlp/lda-vs-pca) - [Stock Regression](https://github.com/pparkitn/kaggle/blob/main/us-stock-market-prediction-histdecision-trees.ipynb) · [Kaggle](https://www.kaggle.com/code/dsptlp/regression-with-timeline-data) - [Random Forest + SHAP](https://github.com/pparkitn/kaggle/blob/main/classification-random-forest.ipynb) · [Kaggle](https://www.kaggle.com/code/dsptlp/classification-random-forest)Resources
Technical notes: GPU/CUDA setup · Spark patterns · AWS · Kaggle · Papermill
Education
- MIDS — UC Berkeley · Statistical ML, Deep Learning, Systems at Scale
- B.Sc. Eng Systems & Computing — University of Guelph · Control Systems, Signal Processing, Software Engineering