Skip to content
View Kenchch's full-sized avatar
  • Christchurch

Highlights

  • Pro

Block or report Kenchch

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Kenchch/README.md

Feng Jiang

Data and analytics engineering — Python, SQL, Spark, Airflow and dbt.

Based in Christchurch, New Zealand, and open to data and analytics engineering roles. Contact.

I build pipelines with explicit data contracts, recoverable publication and reconciled outputs. This portfolio also covers statistical analysis and measured ML deployment trade-offs.

Selected engineering work

Project Implementation and evidence Scope and trade-offs
retail-ai-pipeline Airflow orchestration, versioned publication through an atomic manifest, a contracted dbt/DuckDB mart and a Power BI semantic model. 522,566 loaded + 19,343 quarantined = 541,909 input rows. Single-machine storage and execution. Warehouse and downstream mart promotion are separate boundaries; adoption telemetry is simulated.
nz-attraction-pageviews Incremental Wikimedia ingestion into DuckDB: retries, quarantine, watermark recovery and an offline demo. Handles ambiguous missing responses and publication lag; pageviews measure online attention, not attendance. See limits.
aerial-small-object-detection YOLO11n on VisDrone2019, ONNX export and accuracy parity checks, CUDA placement verification, and separate core/transfer-inclusive latency measurements. Validation-based checkpoint selection and a limited training budget. Synthetic-motion tracking demo demonstrates the pipeline, not tracking accuracy on moving objects.

Analysis and distributed processing

Project Technical focus Evidence
online-retail-analysis-r R/SQLite analysis with chronological, one-to-one credit matching, SHA-256 input verification and checks against committed summary tables. Rendered analysis, including the published correction and exact-match limitations.
nz-sheep-decline-by-region Stats NZ source reconciliation, suppression-aware comparisons and sensitivity to the choice of observation window; R + Quarto. Published report, with validation rules exercised against deliberately corrupted data.
Million Song PySpark genre models and implicit-feedback ALS; training-only resampling, ranking-metric tests and real-Spark model smoke tests. Recorded study results and provenance. Historical results are distinct from the current CLI and synthetic CI fixtures.
GHCN-Daily Fixed-width metadata parsing, station enrichment and weather aggregation in Spark; synthetic regression tests exercise quality flags and unit conversion. Workflow and interpretation limits. Station summaries are not coverage-adjusted national climate estimates; CI does not establish archive-scale performance.

Reading the results

These are portfolio and study projects, not evidence of production deployment or realised commercial impact. Dataset revenue totals and offline model benchmarks are reported with their respective definitions and limitations.

The two retail projects use the same UCI source but different acceptance and cancellation rules. Their READMEs reconcile gross positive sales with the R analysis's matched-cancellation result; partial and unmatched returns remain outside that netting method.

Pinned Loading

  1. aerial-small-object-detection aerial-small-object-detection Public

    YOLO11 small-object detection on VisDrone2019: an accuracy/latency study of input resolution, with ONNX export and backend benchmarking.

    Python

  2. nz-attraction-pageviews nz-attraction-pageviews Public

    Incremental API ingest of Wikipedia pageviews for NZ visitor attractions, into DuckDB

    Python

  3. online-retail-analysis-r online-retail-analysis-r Public

    Reproducible R and SQL analysis of 541,909 retail invoice lines, with cancellation-aware netting and an auditable £9.88M result.

    R

  4. retail-ai-pipeline retail-ai-pipeline Public

    Retail data pipeline with data-quality checks, Airflow/dbt orchestration, Power BI reports and product recommendations. Portfolio demonstration.

    Python 1