DMML -- Question Bank -- BITS Subject Vault -- cvam.sight

Questions extracted and expanded from the five uploaded decks in Slides/: CS1 Data Representations, CS2 Data Management Fundamentals, L3 Data Architectures, L4 Data Pipelines, and L5 Modern Data Infrastructure and DataOps. This page intentionally has no answers. Work from memory first, then check the slides explained and cheatsheet.

How to use

Tag	What it tests
define	Clean definitions and vocabulary.
compare	Side-by-side differences and tradeoffs.
scenario	Apply a concept to a realistic data/ML system.
design	Choose architecture, pipeline or platform pattern.
gotcha	Common mistake, edge case or operational failure mode.

CS1 -- Data Representations

define Define data, information, knowledge and wisdom. Give one example that moves through all four levels.
define What is data management, and which verbs in the definition are operationally important?
scenario A dashboard says the pipeline succeeded, but an ML model suddenly predicts badly for one country. Which data-management concepts explain this failure?
compare Compare structured, semi-structured, unstructured and quasi-structured data with examples.
compare Compare relational, hierarchical, graph, document and key-value data models. For each, name one strong use case.
gotcha Why can a single wide relational table create redundancy, integrity and expansion problems?
compare Explain declarative vs imperative query languages using a non-database example.
compare When is row-major storage better than column-major storage? When is column-major better?
define What is serialization? Why must serializer and deserializer agree on format, schema and algorithm?
compare Compare CSV, JSON, XML, Protobuf and Avro as serialization formats.
compare Compare OLTP and OLAP by workload, data model, latency expectation and query style.
define Explain roll-up, drill-down, slice-and-dice, rotation and trend analysis in OLAP.
compare Compare ACID and BASE. Why do distributed systems often expose this tradeoff?
scenario A product catalog requires fast lookup by SKU, while a recommendation system needs relationship traversal among users/products. Which models fit each problem?

CS2 -- Data Management Fundamentals

define Explain "data as an asset" and "data as a liability" with one ML-system example for each.
gotcha Critique the statement: "More data always improves machine learning."
scenario What legal and operational issues appear when collecting personally identifiable information for a model?
scenario Why can deletion be harder than it sounds in distributed storage, backups and derived datasets?
define What are durability, consistency, versioning/time travel, freshness and availability in data reliability?
define List the core components of a data management strategy and explain why each matters.
compare Compare data integration, data processing, data storage, data governance and data security.
define What is metadata? Give three examples of metadata that help data consumers.
scenario A salary table is needed for payroll analytics, but most employees must not see salaries. Which governance and security controls apply?
define What is a data platform? How is it different from a single database?
design Draw the data life cycle from generation to serving. Where do validation, cleaning, enrichment and storage happen?
compare Compare data analyst, database administrator, data engineer, data scientist and ML engineer roles.
scenario For an ecommerce personalized recommendation engine, list the data sources, cleaning steps, storage choices and delivery path you would inspect first.
compare Explain the conceptual relationship among data architecture, data engineering and data management.

L3 -- Data Architectures

define What is data architecture according to TOGAF-style and DAMA-style thinking?
compare Compare operational and technical aspects of data architecture.
compare Compare centralized, decentralized and hybrid data architectures.
scenario A company has duplicated customer truth across sales, support and marketing. Which architecture pattern helps, and what tradeoffs remain?
define What is a data warehouse schema? Why is it optimized differently from an operational database schema?
compare Compare star schema and snowflake schema using fact and dimension tables.
define What is a data mart? Why might HR or IT want a data mart instead of querying the entire warehouse?
compare Compare data warehouse, data lake and data lakehouse.
gotcha What makes a data lake turn into a data swamp?
define Explain data mesh: domain ownership, data products and self-serve platform.
define Explain data fabric: unified data layer, APIs, CDC and virtualization.
design Design a fraud alert architecture for a bank needing historical reporting and real-time alerts. Which parts are batch, speed and serving?
compare Compare Lambda and Kappa architecture for big data systems.
scenario For ML workflows, when would Lambda be preferable to Kappa, and when would Kappa be simpler?
define List and explain the 5 Vs of big data.
design Name the components of a big data architecture from source to serving and analytics.

L4 -- Data Pipelines

define What problem does a data pipeline solve? Why is manual feeding of source data not enough?
compare Compare data flow through databases, through services and through message passing.
gotcha During a rolling upgrade, newer code writes a new database field and older code reads/writes the same record. What should the old code preserve?
define What is schema evolution? Why is rewriting a large dataset to a new schema often avoided?
compare Compare client-server architecture, SOA and microservices.
compare Compare REST, SOAP and RPC. Why is location transparency dangerous?
define What is a message broker? List three reliability benefits over direct RPC.
compare Compare traditional/legacy data stack and modern data stack.
define List key characteristics of a modern data stack: self-service, discovery, context, lineage, pay-as-you-go and agile data management.
compare Compare modern ingestion, storage/processing, transformation, catalog/governance and privacy/access tools.
define Define a data pipeline using source, rules, transformation, processing and destination.
compare Compare ETL and ELT. Which one fits a legacy warehouse with limited compute? Which one fits a scalable cloud warehouse?
compare Compare batch processing, stream processing and change data capture.
scenario An ecommerce stock database must keep a replica current for failover. How would CDC help?
define Explain the Dataflow model and why unified batch/streaming matters.
design Break a complex pipeline into micro-pipeline stages. What should be independently testable?
scenario Choose tools for a pipeline that starts with SaaS data, lands in a warehouse, transforms with SQL, and powers dashboards.

L5 -- Modern Data Infrastructure and DataOps

define What are the stages of data pipeline architecture from generation/source to serving?
define List the major drivers that changed data infrastructure in the last two decades.
compare Compare ingestion from application databases, REST APIs, Kafka/streams and cloud buckets.
define What does data transformation include beyond "format conversion"?
scenario Why does pipeline complexity eventually require workflow orchestration?
define List raw ingredients of storage infrastructure: HDD/SSD/RAM, networking/CPU, serialization, compression and caching.
compare Compare single-machine and distributed storage.
compare Compare strong consistency and eventual consistency in distributed storage.
compare Compare file storage, block storage, object storage, memory cache, HDFS and streaming storage.
gotcha Why does object storage not support random writes or append in the same way as a local file system?
define What are data storage abstractions? Name warehouse, lake, lakehouse and platform as abstractions.
define What is reverse ETL? Why is sending model scores back to CRM often better than emailing a spreadsheet?
compare Compare business analytics, operational analytics and embedded analytics.
design Draw the common ML/data science infrastructure stack: data warehouse, compute resources, scheduler, architecture, versioning, model ops, feature engineering and model development.
define What responsibilities do data scientists and ML engineers gain when infrastructure is self-service?
compare Compare file exchange, databases, streaming systems, query federation and notebooks as ways to serve data.
define What is infrastructure automation? Explain IaC, GitOps and orchestrated pipeline automation.
compare Compare DevOps and DataOps. What changes when the "artifact" is a data pipeline and not just code?
define Explain data at rest, data in transit and data in use.
design Explain CT/CD for ML: experimental phase, continuous integration, automated training, validation, deployment and monitoring.
scenario A DoorDash-style ML platform has fragmented workflows, low observability, data drift and slow deployment. Which components reduce those problems?

Long/design exercises

design Design a data platform for an ecommerce recommendation engine. Include source systems, ingestion, validation, storage, transformation, feature serving, model training, model serving, monitoring and reverse ETL.
design Design a real-time fraud detection platform. Compare a Lambda design and a Kappa design for the same problem.
design A hospital wants shared data across doctor app, pharmacist app and patient app. Discuss database integration, service integration and message broker integration.
scenario A model fails after a region's data disappears but the data pipeline still completes. Diagnose the failure using data sensitivity, data quality, monitoring and ML observability.
scenario A data lake contains years of raw files but analysts do not trust it. Propose governance, metadata, catalog, quality and access-control fixes.
compare Write a full comparison of warehouse, lake, lakehouse, mesh and fabric. Include when each pattern fails.
design Build a modern data stack from scratch for a startup with SaaS product analytics, Postgres app DB, Kafka events, warehouse analytics and ML training.
gotcha Explain why "pipeline success" is weaker than "data product correctness." What extra signals should be monitored?

Practice rule. Answer in this order: definition -> diagram or comparison -> example -> tradeoff/gotcha. DMML questions usually reward architecture judgement, not memorized tool names.

DMML question bank -- questions only, no answers.