Cutting the First Pattern: What a Feature Actually Is
why a tailor never sells a customer the bolt of fabric itself, and how a feature does the same work of cutting raw data down to a model's exact measurements.
Part 1
Part 2
The Fitting Date That Matters: Point-in-Time Correctness
why a tailor can only use the measurements taken at the actual fitting appointment, and what goes wrong when a pattern is quietly cut using numbers from after the fact.
Part 3
Made-to-Order or Off the Rack: Online and Offline Features
why some garments are cut fresh the moment a customer walks in and others are pulled ready-made from the rack, and how features split the same way between live and batch.
Part 4
The Alteration Ticket: Feature Versioning and Lineage
why every alteration made to a garment gets logged on a ticket pinned to the original order, and why a feature needs the same traceable record of every change made to it.
Part 5
From Bolt to Garment: Feature Transformation Pipelines
why cutting a pattern is only the first of many steps between raw cloth and a finished garment, and how a feature pipeline strings together the same kind of sequence.
Part 6
Measured Over Many Fittings: Aggregation Windows
why a good tailor tracks how a client's measurements have trended across several fittings rather than trusting any single one, and how aggregation windows do the same for features.
Part 7
The Swatch Book Code: Encoding Categorical Features
why a tailor's swatch book assigns every fabric a reference number rather than describing it in prose each time, and how encoding does the same for categorical data feeding a model.
Part 8
One Fitting Room for the Whole House: Feature Stores as Shared Infrastructure
why a tailoring house keeps one shared fitting room and pattern archive instead of letting every tailor build their own from scratch, and why feature stores exist for the same reason.
Part 9
The Same Tape Measure Twice: Training and Serving Skew
why a garment cut from one measurement and altered against a slightly different one never quite fits, and why a model trained on one version of a feature and served another behaves the same way.
Part 10
The Rush Alteration: Freshness and Latency for Online Serving
why a same-day alteration has to happen in minutes while a full bespoke order can take weeks, and how online feature serving lives under the same unforgiving clock.
Part 11
Re-Measuring the Archive: Backfilling Historical Features
why a tailoring house sometimes has to reconstruct a client's measurements from years of old order slips, and why training a new model on history means backfilling features the same careful way.
Part 12
When the Customer's Measurements Change: Monitoring Feature Drift
why a good tailor re-measures a regular client rather than assuming last year's numbers still hold, and why a feature needs the same ongoing watch for quiet, gradual change.
Part 13
One Pattern Block, Many Ateliers: Feature Reuse Across Teams
why a proven pattern block gets shared across every workroom in a tailoring house instead of being redrawn from scratch each time, and why the best features work the same way across teams.
Part 14
Searching the Swatch Book: Feature Discovery
why a well-organized swatch book needs more than shelves to be useful, and why a growing catalog of features needs genuine search, not just a place to store them.
Part 15
The Apprentice Who Drafts Patterns: Automated Feature Engineering
why a capable apprentice can draft a first-pass pattern from a client's measurements without being told exactly how, and why AI can now do something similar for candidate features.
Part 16
Grading the Seams: Feature Quality Scoring
why a master tailor inspects every seam against a clear standard before a garment leaves the shop, and why a feature needs the same disciplined quality check before it reaches a model.
Part 17
A Sketch on a Napkin, Cut Into Cloth: LLM-Generated Features from Unstructured Data
why a talented tailor can turn a client's rough napkin sketch and a few spoken words into an actual cutting pattern, and why large language models can now do something similar with unstructured text.
Part 18
The Cost of Bespoke: Pricing Feature Computation
why a fully bespoke suit costs more than a rack adjustment, and why every feature in a model carries its own real, ongoing computational price tag worth knowing.
Part 19
When You Don't Need a Whole Atelier: Feature Stores Are Overkill
why a small neighborhood tailor shop with two people and a handful of regular clients has no need for an industrial pattern warehouse, and why plenty of ML teams don't need a full feature store either.
Part 20
The Finished Collection: Every Pattern, One Well-Dressed System
the first pattern cut, the fitting date honored, the shared fitting room, the rush alteration, and the apprentice's drafted sketch, every article's lesson reassembled into one finished collection.