← Back to portfolio
Selected production work

Ownership beyond
the model.

Selected systems I build and own: credit decisioning, document intelligence, analytics infrastructure, and entity resolution.

01 / Cashco Financial · 2024–present

Credit decisioning,
from design to operation.

My role: architect and end-to-end technical owner.
From financial-data ingestion and model development to production delivery, evaluation, and ongoing operation.

I architected and own a production decisioning system that combines financial-data ingestion, transaction and behavioral feature engineering, calibrated ML scoring, explicit policy orchestration, explainability, versioning, and production monitoring.

Modeling

Productionized routed, calibrated risk models and engineered features from financial transactions and behavior. Evaluate discrimination and probability quality, and use champion/challenger evaluation to assess changes.

Decisioning

Keep model estimation separate from policy decisions, with explicit review paths and safe fallbacks. Integrate SHAP-based explanations and reason codes with versioned decision evidence to support review and investigation.

Production

Build and operate Python/FastAPI services using Docker and Azure. CI/CD, regression validation, versioned releases, and structured logging support controlled delivery and auditable decisions.

Lifecycle

Use monitoring, shadow evaluation, and outcome feedback to guide model and policy iteration. Connect investigation of input quality and model behavior with the next round of evaluation and engineering work.

The architecture overview explains my engineering responsibilities through a conceptual flow. The linked labs are independent synthetic demonstrations; they do not reproduce employer configurations, decision rules, or performance results.

02 / Cashco Financial · Document intelligence

Make unstructured data
useful and reviewable.

My role: design and implementation.
Document extraction, validation, and quality gates for financial inputs.

My contribution

I build document-processing workflows that turn unstructured financial inputs into usable data. Extraction is paired with explicit quality gates, financial-consistency checks, and review paths for uncertain or incomplete evidence.

Explore the general methods

My independent public labs illustrate extraction evaluation, data-quality checks, and transaction feature engineering using synthetic inputs. They make the methods inspectable without presenting employer data or operational results.

03 / Cashco Financial · Data engineering

Build the foundations
for useful analytics.

My role: platform development and operation.
Airflow orchestration, dbt modeling, data quality, and Metabase reporting.

My contribution

I architected an analytics platform using Airflow, dbt, and Metabase. My ownership spans ingestion, orchestration, analytical modeling, data quality, reporting, and ongoing operation.

The engineering work connects maintainable pipelines with curated data models and reporting that business teams can use and review.

Public engineering examples

The accompanying repository contains generic Airflow, dbt, PostgreSQL, and reverse-proxy examples. Its design notes discuss common trade-offs in ingestion, data quality, and BI engineering.

These are illustrative materials informed by professional experience. They do not document an employer’s current deployment, infrastructure capacity, performance, costs, or security posture.

04 / Cybera · Applied NLP

Entity resolution
at distributed scale.

Connect unstructured records with learned representations, retrieval, and reproducible distributed execution.

What I built

Designed and deployed an entity-resolution pipeline for unstructured data using transformer embeddings, Siamese neural networks, and retrieval. GPT-assisted labeling and normalization reduced manual preparation.

Execution and evaluation

Scaled execution with Docker, multiprocessing, GPU acceleration, and a 24-node GCP virtual-machine cluster. The reported classification accuracy was 92% on the project evaluation.

BERT / LLaMA2 FAISS TensorFlow GCP

Results are summarized from my professional experience. Employer source code and customer data are not published. Selected earlier collaborative geospatial work is linked from the research page.

Start a conversation

Building a team that takes ML
all the way to production?

I’m interested in senior data science and ML engineering roles with hands-on technical ownership. Based in Calgary, Canada.