OACintermediate
OAC Datasets: Preparation, Joins & Data Enrichment
Explain how OAC datasets are prepared, cleaned, joined, and enriched before being used in Data Visualization projects.
A dataset usually needs shaping before it is useful. OAC's data preparation handles that without leaving the tool.
It's like a professional kitchen's prep station: instead of chopping vegetables from scratch every single time an order comes in, the recipe (transformation steps) is defined once, and every future batch of the same raw ingredient gets automatically prepped the same consistent way before it ever reaches the chef.
Key Concepts
1
Profiling happens on load. OAC inspects each column and infers whether it is a measure, an attribute or
a date, and shows the distribution — nulls, outliers, inconsistent formats.
2
That profile is the useful part. It tells you the file has 4% nulls in a key column before you build
anything on it.
3
The preparation script is a recorded sequence of steps.
rename a column
change a type
split "Bengaluru, KA" into city and state
filter out the header rows the export included
create a calculated column4
Each step is recorded, so the script re-runs when the source refreshes. That is the difference from
editing a spreadsheet: the cleaning is repeatable rather than done once by hand.
5
Joins between datasets are defined in the dataset itself, so a project sees one combined source rather
than performing the join every time.
6
Enrichment is the feature people do not expect. OAC recognises the semantics of a column and offers
additions:
7
a postcode column -> offers city, state, latitude, longitude
a company name -> offers industry classification8
Accepting one adds a column derived from reference data, which is how a map visualisation becomes possible
from a file that only had postcodes.
9
Where this fits, and the caution. Preparation is the self-service end of the pipeline: an analyst
shapes a file this afternoon rather than raising a request.
10
The caution is that a preparation script is owned by whoever built it and lives inside OAC. When the same
cleaning is needed by several projects, or when it encodes business rules, it belongs upstream in the
warehouse — otherwise the organisation has two versions of "clean" and no way to know which a number came
from.
11
The practical signal. A preparation script that has grown to twenty steps and is copied between projects
is telling you the work has moved past self-service and should be done once, in the pipeline.
12
What the interviewer is probing.1. "What can you do in dataset preparation?" Probing: the capability. Stalls: "Load data."
Moves up: profile columns, fix types, split and concatenate, handle nulls and apply recommended
enrichments, saved as repeatable steps.
13
2. "What is auto-enrichment?" Probing: the feature. Stalls: "Cleaning." Moves up: the
system recognising a column such as a postcode and offering derived attributes like region, which it
joins from reference data.
14
3. "What is the risk with self-service joins?" Probing: the honest point. Stalls: "None."
Moves up: a wrong grain or a many-to-many join silently inflates measures, and there is no
modeller reviewing it.
15
4. "When should a dataset be promoted into the semantic model?" Probing: the governance.
Stalls: "Never." Moves up: once several people depend on it — that is the point at which it
needs an owner, a definition and security.