Research through real operations
Client work gives me the questions worth investigating: forecasting demand, reducing waste, and making recommendations clear enough that people can challenge them before acting.
The work creates the question. The writing makes the answer travel.
My writing is not a separate content stream. It begins inside the systems I build, follows the evidence through applied machine learning, and returns as something other people can inspect, use, or challenge.
Technical findings for TDS
Towards Data Science is where an experiment earns its technical explanation: the baseline, the method, the failures, the metrics, and what the evidence actually supports.
The human story on Medium
Medium holds the wider journey behind the work: how my questions change, what building with real people teaches me, and how engineering is reshaping the way I think.
One body of work.
Five connected stages.
Observe the operation
Start with a repeated decision inside real client work.
Frame the ML question
Define the target, evidence, constraints, and honest baseline.
Test and evaluate
Compare methods, document failures, and measure useful outcomes.
Publish the finding
Turn the technical result into a rigorous TDS article.
Reflect and connect
Write the builder story on Medium and preserve the case study here.
Research in progress,
published foundations.
Decision intelligence that shows its work
Azumie and the Senza Fine engagement are becoming the practical ground for research into forecasting, operational recommendations, and trustworthy decision support.
Read the Azumie case studyRecommender systems, from formulation to filtering
An early long-form explanation of content-based recommendation, collaborative filtering, low-rank matrix factorization, and mean normalization.
Read on MediumDealing with large datasets
One of my early Towards Data Science pieces, preserved as evidence of where my public learning began and the technical voice I am now returning to.
Read the TDS articleFOLLOW THE WORKTechnical depth.