The Data Warehousing Interview Masterclass
300+ Questions, Real‑World Scenarios & Labs – From On‑Prem to Cloud AI
🧠 The Data Warehouse Visionary Mindset (10 Q&As)
Q1: What is the single most important principle when designing a data warehouse? Most Expert
"Deliver business value, not just data. Every table, every aggregation must answer a real business question. I always start with the CEO's top 5 KPIs and work backwards to the data model – that's how I turned a chaotic reporting system into a trusted decision engine."
Q2: How do you explain a data warehouse to a non‑technical executive? Beginner
"I compare it to a library: the operational systems are like thousands of scattered sticky notes; the data warehouse is a well‑organised library where every book (fact) is indexed, categorised, and instantly retrievable. It gives leaders one version of the truth."
Q3–Q10: Quick mindset gems Intermediate
Q3: Why does data governance matter in DW? Garbage in, garbage out – governance ensures trust.
Q4: What's the difference between a database and a data warehouse? DB for transactions (OLTP), DW for analytics (OLAP).
Q5: How do you align DW with business strategy? Map every dimension and fact to a corporate objective.
Q6: Why is data lineage critical? So you know exactly where a number came from – builds confidence.
Q7: Describe a time you influenced a business decision with DW data. I revealed that 40% of returns were due to late deliveries, prompting a logistics overhaul.
Q8: What is the role of a data steward? Owns data quality and definition for a domain.
Q9: How do you foster a data‑driven culture? Make self‑service easy, provide training, and celebrate data wins.
Q10: What is the future of data warehousing? AI‑augmented warehouses that self‑optimise and answer natural language questions.
🌱 Beginner Level – Data Warehousing Foundations (80+ Q&As)
Q11: What is a data warehouse? Beginner
A centralised repository that stores integrated data from multiple sources, optimised for query and analysis. It supports historical reporting and decision‑making.
Q12: What are the key characteristics of a data warehouse? Beginner
Subject‑oriented, integrated, time‑variant, and non‑volatile (Inmon’s definition). I memorise it with "SITN".
Q13: Explain OLTP vs OLAP. Beginner
OLTP (Online Transaction Processing) handles day‑to‑day operations, fast inserts/updates, normalised. OLAP (Online Analytical Processing) is for complex queries, aggregated reads, denormalised for speed.
Q14: What is a star schema? Beginner
A central fact table surrounded by dimension tables, like a star. Fact contains measures (sales amount), dimensions contain descriptive attributes (product, time). It’s simple and fast.
Q15: Draw a simple star schema for retail sales. Beginner
Fact_Sales: DateKey, ProductKey, StoreKey, Quantity, Amount Dim_Date: DateKey, Day, Month, Year Dim_Product: ProductKey, Name, Category Dim_Store: StoreKey, City, Region
Q16: What is a fact table? Types of facts? Beginner
Contains measurable business events. Types: additive (sales – can sum across dimensions), semi‑additive (inventory – sum across some), non‑additive (ratios, percentages).
Q17: What is a dimension table? Provide examples. Beginner
Descriptive reference data: Customer, Product, Time, Store. It provides context for measures.
Q18: What is a surrogate key and why use it? Beginner
An artificial integer key replacing natural keys. It insulates the warehouse from source system changes, improves join performance, and handles slowly changing dimensions gracefully.
Q19: What is a slowly changing dimension (SCD)? Beginner
Dimension attributes that change over time. We handle them with SCD Type 1 (overwrite), Type 2 (add new row), Type 3 (add new column).
Q20: How do you implement SCD Type 2 in a table? Beginner
Add Start_Date, End_Date, and a Current_Flag. When a change occurs, expire the old row and insert a new one with updated values.
Q21: What is ETL? Beginner
Extract, Transform, Load – the process of moving data from source systems, cleaning and transforming it, then loading into the warehouse.
Q22: ETL vs ELT – explain the difference. Beginner
ETL transforms before loading (traditional). ELT loads raw data first, then transforms inside the warehouse using its compute power (modern cloud approach). I use ELT with BigQuery and dbt for flexibility.
Q23: What is a staging area? Beginner
An intermediate storage where raw data lands before transformation. It allows auditing, reprocessing, and decouples extraction from transformation.
Q24: What is a data mart? Beginner
A subset of the data warehouse focused on a specific business line (e.g., sales mart, marketing mart). It improves query performance and access control.
Q25: Kimball vs Inmon methodologies – key differences. Beginner
Kimball: bottom‑up, star schemas, dimensional modelling first. Inmon: top‑down, normalised 3NF enterprise data warehouse, then dimensional marts. I choose based on organisation size and agility needs.
Q26: What is a conformed dimension? Beginner
A dimension shared across multiple fact tables (e.g., Date dimension). It ensures consistency and enables cross‑business‑process analysis.
Q27: What is a junk dimension? Beginner
A dimension that combines low‑cardinality flags and attributes (like order status, payment type) to avoid cluttering the fact table.
Q28: What is a degenerate dimension? Beginner
A dimension key stored in the fact table without a corresponding dimension table (e.g., transaction ID). Used for operational tracking.
Q29: How do you design a date dimension? Beginner
Include columns: DateKey (YYYYMMDD), Day, Month, Quarter, Year, DayOfWeek, HolidayFlag. I generate it once for 20 years and it never changes.
Q30: What is a measure? Give additive vs non‑additive examples. Beginner
Measure = numeric business metric. Additive: sales amount (SUM). Non‑additive: unit price (can’t sum), ratios. I always flag the aggregation type in the metadata.
Q31: What are the three types of data warehouse systems? Beginner
1) Enterprise Data Warehouse (EDW), 2) Operational Data Store (ODS) for near real‑time operational reporting, 3) Data Mart.
Q32: What is a view? How is it used in DW? Beginner
A saved SQL query that acts like a virtual table. I use views to simplify complex joins for business users and to implement row‑level security.
Q33: Explain the concept of grain in a fact table. Beginner
The granularity – the most atomic level of each row. For example, "one row per transaction line item". Getting the grain right is the most critical design decision.
Q34: What is a snowflake schema? Pros and cons. Beginner
Normalised dimensions (e.g., Product → SubCategory → Category). Pros: saves space. Cons: more complex queries, slower performance. I avoid it unless storage is a huge concern.
Q35: What is a role‑playing dimension? Beginner
The same dimension used in multiple ways (e.g., Date used as OrderDate, ShipDate). I create views or synonyms to give them distinct names.
Q36: What is a factless fact table? Beginner
A fact table with no measures, only foreign keys. Used to track events like "student attended course" (attendance). It answers questions about conditions or coverage.
Q37: What is a bridge table? When is it used? Beginner
To handle many‑to‑many relationships between a fact and a dimension (e.g., a patient with multiple diagnoses). It's a secondary dimension linking fact to multi‑valued attributes.
Q38: What is the difference between a full load and incremental load? Beginner
Full load replaces the entire table. Incremental loads only new/changed records using timestamps or CDC. I use incremental for large fact tables to reduce load windows.
Q39: What is CDC (Change Data Capture)? Beginner
A technique to identify and capture changes in source data. I use SQL Server CDC or Debezium for real‑time streams into the warehouse.
Q40: How do you handle late arriving dimensions? Beginner
Insert a placeholder dimension row with the natural key, and update it when the real attributes arrive later. The fact loads on time.
Q41: What is a snapshot fact table? Beginner
Captures the state of a process at regular intervals (e.g., daily inventory levels). I use them for trend analysis over time.
Q42: Define data warehouse architecture layers. Beginner
Typically: Source systems → Staging → Integration (EDW) → Presentation (data marts) → BI tools. I also add a semantic layer.
Q43: What is a metadata repository? Beginner
Stores technical and business metadata about the warehouse – table definitions, lineage, ownership. It's the "data about data".
Q44: What is an aggregate table? Why use it? Beginner
Pre‑calculated summary tables (e.g., daily sales by product). They dramatically speed up common queries. I use query rewrite or tools like OLAP cubes to hit aggregates automatically.
Q45: How do you ensure data quality in a data warehouse? Beginner
Proactive: validate at source, clean during ETL. Reactive: data profiling, anomaly detection, reconciliation checks. I implement data quality dashboards with business rules.
Q46–Q90: 45 more beginner essentials (abbreviated) Beginner
Q46: What is a bitmap index? Good for low‑cardinality columns, used in OLAP (Oracle).
Q47: What is partitioning? Dividing large tables for manageability and performance.
Q48: List partitioning vs range partitioning. List by discrete values (region), range by intervals (date).
Q49: What is a materialized view? Stored result of a query refreshed on demand – improves performance.
Q50: What is a columnar storage? Data stored by column – ideal for analytical queries (Parquet, ORC).
Q51: Name three popular on‑prem data warehouse platforms. Teradata, Oracle Exadata, SQL Server.
Q52: What is SQL Server Analysis Services (SSAS)? OLAP cube engine for multidimensional analysis.
Q53: What is Teradata's primary index? Determines data distribution across AMPs.
Q54: What is Netezza? IBM’s data warehouse appliance – fast, FPGA‑accelerated.
Q55: What is Greenplum? Open‑source MPP database based on PostgreSQL.
Q56: Explain a MPP (Massively Parallel Processing) architecture. Distributes data and queries across many nodes for speed.
Q57: What is a shared‑nothing architecture? Each node has its own CPU, memory, storage – like Teradata.
Q58: What is the difference between a database and a schema? Schema is a container for objects within a database (SQL Server, PostgreSQL).
Q59: What is a tablespace? A logical storage unit to manage physical placement of data.
Q60: What is a fact constellation? Multiple fact tables sharing dimensions (Galaxy schema).
Q61: What is a mini‑dimension? Splitting a large dimension’s rapidly changing attributes into a separate table.
Q62: What is a busy dimension? A dimension with attributes frequently queried together – keep them denormalised.
Q63: What is a heterogeneous source? Data coming from different types of systems – RDBMS, files, APIs.
Q64: What is the purpose of a surrogate key pipeline? To generate and map surrogate keys consistently across ETL.
Q65: How do you handle historical data in a warehouse? SCDs and periodic snapshots.
Q66: What is a rollback segment? Not typical in DW, but important in OLTP for transaction consistency.
Q67: What is a data dictionary? A documentation of all data elements – critical for onboarding.
Q68: How to rename a column in a fact table? Add new column, dual‑load both for a period, then drop old after consumers migrated.
Q69: What is a heap table? A table with no clustered index – log‑structured.
Q70: What is a B‑tree index? Balanced tree structure for fast lookups – common in OLTP, but in DW we prefer columnar.
Q71: What is a star transformation? A query optimization technique in Oracle for star schemas.
Q72: What is the difference between a local and global index? Local partitioned index, global covers entire table.
Q73: Explain the use of a calendar table. A date dimension with fiscal periods, holidays – join anchor.
Q74: What is a check constraint? Enforces a condition on column values – I use it for data integrity in DW.
Q75: What is the purpose of a primary key in a dimension? Uniquely identifies each row; should be surrogate.
Q76: How do you handle NULLs in a dimension? Use descriptive values like 'Unknown', 'N/A' to avoid NULL join issues.
Q77: What is a dimension outrigger? A secondary dimension attached to a dimension (snowflake variation).
Q78: What is a fact table grain "transaction line item"? One row per product on an order.
Q79: How to model a hierarchy (region → country → city)? Either flatten into one dim or use a bridge for ragged hierarchies.
Q80: What is a Type 2 SCD with a history table? The dimension table itself stores all history.
Q81: What is a Type 1 SCD example? Update product category name – no history tracked.
Q82: What is a Type 3 SCD? Adds a "previous value" column – limited history.
Q83: What is a "current flag" in SCD Type 2? A column indicating the latest row (1) vs historical (0).
Q84: How to handle very large dimensions (millions of rows)? Use mini‑dimensions for frequently changing attributes, keep static ones in main dim.
Q85: What is a "conformed fact"? Facts using conformed dimensions – allows cross‑fact analysis.
Q86: What is a "drill‑across" query? Query that combines multiple facts using conformed dimensions.
Q87: What is a "semi‑additive fact" example? Account balance – sum across accounts but not dates.
Q88: What is a "periodic snapshot" fact table? Captures state at regular times (e.g., month‑end inventory).
Q89: What is an "accumulating snapshot" fact table? Tracks a process with defined milestones (order processing).
Q90: What is a "transaction fact" table? Most common – one row per event, additive.
🚀 Intermediate Level – Modeling, ETL & Performance (90+ Q&As)
Q91: How do you optimize a slow‑running aggregation query on a 1 billion row fact table? Intermediate
"I first check if aggregate tables or materialized views exist. I review partitioning strategy (date), ensure columnar storage, and check for statistics staleness. If ad‑hoc, I create a covering index. Once, I reduced a 30‑min query to 3 seconds by switching to a date‑partitioned table and using a pre‑aggregated summary."
Q92: What is a data vault model? When would you use it? Intermediate
"A hybrid approach using Hubs (business keys), Links (relationships), and Satellites (descriptive attributes). It's agile, scalable, and handles change well. I led a data vault implementation for a healthcare provider – it absorbed 10+ source system changes without re‑engineering."
Q93: How do you implement ETL for a slowly changing dimension Type 2 in SQL? Intermediate
-- Expire old records
UPDATE dim_customer SET end_date = GETDATE(), is_current = 0
WHERE customer_id = ? AND is_current = 1;
-- Insert new row
INSERT INTO dim_customer (customer_id, name, start_date, end_date, is_current)
VALUES (?, ?, GETDATE(), '9999-12-31', 1);Q94: What is a surrogate key pipeline? How do you build one? Intermediate
"A process that maps natural keys to integer surrogate keys. I use a lookup table and a sequence generator. In modern ELT, I generate surrogate keys using HASH functions or identity columns in the staging layer."
Q95: What is a fact table grain? Why is it critical? Intermediate
"Grain defines the level of detail for each row. Getting it wrong means you can't answer the business questions. I always co‑design grain with the business owner: 'one row per sales order line per day' – that clarity saved a project from disaster."
Q96: Explain a Type 2 SCD implementation with a MERGE statement. Intermediate
MERGE dim_product AS target
USING stg_product AS source
ON target.product_bk = source.product_bk AND target.is_current = 1
WHEN MATCHED AND (target.name <> source.name) THEN UPDATE SET is_current = 0, end_date = GETDATE()
WHEN NOT MATCHED THEN INSERT ...;Q97: What is a bridge table? Provide a real example. Intermediate
"For a patient‑diagnosis scenario, a bridge table PatientDiagnosis linking fact to multiple diagnosis codes. I weight it to allocate measures properly – a $1000 claim might split 50/50 if two primary diagnoses."
Q98: What are semi‑additive facts? How do you handle them in a query? Intermediate
"Inventory quantity is semi‑additive – you can sum across products but not across time. You must use LAST_VALUE or AVG over time. I enforce business rules in the BI layer."
Q99: What is a factless fact table? Design one for student attendance. Intermediate
"Fact_Attendance: DateKey, StudentKey, CourseKey. No measures; the presence of a row signifies attendance. Count(*) gives number of attendees."
Q100: How do you model a multi‑valued dimension (e.g., a customer with multiple hobbies)? Intermediate
"I'd create a bridge table at the customer grain, or for low cardinality, a junk dimension with concatenated flags. Never flatten into repeating columns."
Q101: What is the difference between a star schema and a galaxy schema? Intermediate
"Galaxy (fact constellation) has multiple fact tables sharing dimensions. It’s a more realistic enterprise model."
Q102: How do you handle rapidly changing dimensions? Intermediate
"Split into a main dimension (static attributes) and a mini‑dimension (rapidly changing attributes like demographics). This prevents excessive Type 2 rows."
Q103: What is a degenerate dimension? Give an example from e‑commerce. Intermediate
"Order number. It's a dimension key without its own table – stored directly in the fact. I use it for operational drill‑through."
Q104: What is a role‑playing dimension? Show how to use it in a query. Intermediate
SELECT d_order.Date, d_ship.Date
FROM fact_sales f
JOIN dim_date d_order ON f.order_date_key = d_order.DateKey
JOIN dim_date d_ship ON f.ship_date_key = d_ship.DateKey;Q105: What are the different types of fact tables? Intermediate
Transactional, Periodic Snapshot, Accumulating Snapshot, Factless. I choose based on business process measurement needs.
Q106: How do you design an accumulating snapshot fact table? Intermediate
"Columns for each milestone date (order received, shipped, delivered). A single row is updated as the process progresses. I use it for order‑fulfillment pipelines."
Q107: What is a data warehouse bus matrix? Intermediate
"A Kimball tool mapping business processes to dimensions – ensures common dimensions are identified for conformed design."
Q108: How to estimate data warehouse size? Intermediate
"I calculate expected rows per fact table per day, multiply by column sizes, factor in indexes and compression, then project for 3‑5 years. Usually add 30% buffer."
Q109: What is a clustered columnstore index? Intermediate
"SQL Server’s columnar storage technology – it compresses and accelerates analytical queries. I use it on all large fact tables."
Q110: What are the benefits of using surrogate keys in a data warehouse? Intermediate
"Performance (integer joins), independence from source changes, handling of historical vs current, and ability to integrate multiple sources."
Q111: Explain your ETL error handling strategy. Intermediate
"I use a try‑catch pattern, log errors to an error dimension with source row, timestamp, and error message. Failed rows go to a suspense table for reprocessing."
Q112: How do you handle schema evolution in ETL? Intermediate
"I use schema‑driven pipelines or tools like dbt that allow incremental model changes. I detect new columns via metadata and add them dynamically."
Q113: What is a data lineage tool? Name two. Intermediate
"Tools that track data from source to report. I've used MANTA and Atlan – they help troubleshoot and audit."
Q114: What is a "late arriving fact"? How do you handle it? Intermediate
"When a fact record comes after the dimension has already been updated with new attributes. I use the surrogate key that was active at the fact’s event date by looking it up in the dimension's history."
Q115: What are the typical indexes used in a data warehouse? Intermediate
"Columnstore for fact tables, bitmap indexes for low‑cardinality columns (Oracle), and B‑tree on dimension surrogate keys. I avoid too many indexes to keep ETL fast."
Q116: How do you decide between a star schema and a data vault? Intermediate
"Star for well‑understood, stable reporting needs with fast queries. Data Vault for enterprise‑wide integration, many source systems, and audit trail requirements."
Q117: What is a persistent staging area (PSA)? Intermediate
"A historical copy of raw source data in the warehouse, enabling full reloads and audit. I always include it in my architecture."
Q118: How do you load data into a data warehouse efficiently? Intermediate
"Use bulk insert APIs (BCP, COPY command in Redshift, BigQuery load), disable indexes during load, partition swap for big tables, and parallel loading."
Q119: What is a virtual data warehouse? Intermediate
"A federation layer that queries data in place without moving it. Tools like Denodo or Presto/Trino create a logical view over multiple sources."
Q120–Q180: 61 Intermediate rapid‑fire Intermediate
Q120: What is a data lake? Raw storage for structured and unstructured data, schema‑on‑read.
Q121: What is a lakehouse? Combines data lake flexibility with warehouse ACID and performance (Delta Lake).
Q122: What is Apache Hive? SQL interface on Hadoop – I used it for batch processing.
Q123: What is Presto/Trino? Distributed SQL query engine for data lakes – fast, federated.
Q124: What is Apache Spark? Unified engine for large‑scale data processing – I use it for ETL.
Q125: What is dbt? Transformation tool using SQL, modular, version‑controlled – my go‑to for ELT.
Q126: What is a medallion architecture? Bronze (raw), Silver (cleaned), Gold (aggregates) layers in a lakehouse.
Q127: What is Delta Lake? Open‑source storage layer with ACID, time travel, and schema enforcement.
Q128: What is Apache Iceberg? High‑performance table format for huge datasets – used in many cloud warehouses.
Q129: What is the difference between managed and external tables in Hive? Managed: Hive controls lifecycle. External: you control.
Q130: What is a partitioning key? Column used to split data into partitions – I choose high cardinality and query filter freq.
Q131: What is bucketing (clustering) in Hive? Further divides partitions for faster joins.
Q132: What is an ORC file format? Optimized Row Columnar – better compression and predicate pushdown.
Q133: What is a Parquet file? Columnar format, widely used – I always store data in Parquet on data lakes.
Q134: Explain the concept of a "data pipeline". A series of steps that ingest, transform, and load data.
Q135: What is Airflow? Workflow orchestration tool – I use it to schedule ETL/ELT jobs.
Q136: What is a DAG in Airflow? Directed Acyclic Graph – defines task dependencies.
Q137: What is a data catalog? Searchable inventory of data assets (Alation, Collibra).
Q138: What is data governance framework? Policies, standards, and roles to ensure data is managed properly.
Q139: What is a data steward? A business person responsible for data quality in their domain.
Q140: What is master data management (MDM)? Creating a single, consistent master record for key entities (customer, product).
Q141: How does cloud data warehousing differ from on‑prem? Elastic scaling, separation of compute/storage, pay‑per‑use.
Q142: What is Amazon Redshift? Cloud MPP data warehouse – columnar, based on PostgreSQL.
Q143: Explain Redshift distribution styles: KEY, ALL, EVEN, AUTO. I choose KEY on join column to collocate data.
Q144: What is a sort key in Redshift? Defines the order of data on disk to enable zone maps – improves performance.
Q145: What is VACUUM in Redshift? Reclaims space and re‑sorts rows. I schedule it regularly.
Q146: What is Google BigQuery? Serverless, highly scalable data warehouse – no infrastructure to manage.
Q147: How does BigQuery store data? Columnar Capacitor format, automatically compressed and encrypted.
Q148: What is a slot in BigQuery? Unit of computational capacity – I monitor slot usage for cost.
Q149: What is BigQuery partitioning? Divide table by date or integer range to reduce query costs.
Q150: What is clustering in BigQuery? Sorts data within partitions to improve filter and join efficiency.
Q151: What is Snowflake? Cloud data platform with separation of storage, compute, and services.
Q152: Explain Snowflake’s architecture: cloud services, virtual warehouses, and storage.
Q153: What is a virtual warehouse in Snowflake? A compute cluster that you can scale up/down and suspend.
Q154: What is zero‑copy cloning in Snowflake? Instant clone without additional storage cost – great for dev/test.
Q155: What is Time Travel in Snowflake? Access historical data up to 90 days – undo mistakes.
Q156: What is data sharing in Snowflake? Securely share live data across accounts without copying.
Q157: What is Azure Synapse Analytics? Unified analytics service combining data warehouse, Spark, and pipelines.
Q158: What is a dedicated SQL pool in Synapse? The MPP engine, previously SQL DW – I use it for large fact tables.
Q159: Explain Synapse’s distribution: hash, round‑robin, replicate. Similar to Redshift.
Q160: What is PolyBase? Virtualizes external data sources within SQL Server/Synapse.
Q161: What is Databricks? Unified analytics platform built on Spark, with Delta Lake – lakehouse architecture.
Q162: What is a Databricks notebook? Interactive collaborative environment – I do exploratory analysis there.
Q163: What is MLflow? Open‑source platform for managing ML lifecycle – I track experiments.
Q164: How do you integrate AI/ML with a data warehouse? Feature store pulls from DW, model training uses historical data, predictions written back.
Q165: What is a feature store? Centralised repository for ML features, ensures consistency between training and serving.
Q166: How to serve model predictions directly in a data warehouse? Use SQL‑invokable functions (BigQuery ML, Redshift ML, Snowpark).
Q167: What is BigQuery ML? Train and execute ML models using SQL directly in BigQuery.
Q168: How to create a linear regression model in BigQuery ML? CREATE MODEL ... OPTIONS(model_type='linear_reg') AS SELECT ...
Q169: What is Redshift ML? Use SQL to invoke SageMaker models for predictions.
Q170: What is Snowpark? Snowflake’s developer framework for Python/Java/Scala – run ML code inside Snowflake.
Q171: What is an ELT pipeline with dbt and Snowflake? Fivetran → Snowflake → dbt transforms, then BI.
Q172: What is a data warehouse automation tool? WhereScape, BimlFlex – accelerate DW development.
Q173: What is a slowly changing dimension Type 4? Separate history table, current dimension keeps only latest.
Q174: What is a Type 6 SCD? Combines Type 1,2,3 – I rarely use it due to complexity.
Q175: How to handle hierarchical dimensions (employee‑manager)? Use a parent‑child bridge or recursive CTE.
Q176: What is a ragged hierarchy? Uneven depth – typical org chart. Model with bridge table.
Q177: What is a parent‑child dimension? Employee dimension with foreign key to itself (manager_id).
Q178: What is a time series fact table? Stores events at regular intervals – I use it for IoT sensor data.
Q179: What is a derived fact? A calculation based on other facts – I materialise it for performance.
Q180: What is a coverage fact table? Factless table showing which products were available in a store on a date – for inventory analysis.
⚡ Expert Level – Advanced Architecture & Cloud DW (80+ Q&As)
Q181: You are designing a multi‑petabyte data warehouse on a cloud platform – how would you approach it? Expert
"I'd choose a lakehouse pattern (Databricks/Delta Lake or Iceberg on S3) for raw storage, then build a gold layer in Snowflake or BigQuery for BI. Use medallion architecture, data contracts, and govern with a catalog. Compute‑storage separation is key. I architected a similar system for a telecom – 50TB ingested daily, queries under 10s."
Q182: Compare Redshift, BigQuery, and Snowflake for a financial reporting workload. Expert
"Redshift: requires active tuning (sort keys, dist keys), good if you have a DBA team. BigQuery: truly serverless, no knobs, great for unpredictable workloads, but costs can spike. Snowflake: best ease of use, data sharing, and governance; compute isolation is excellent. For strict compliance and multiple departments, I'd lean Snowflake."
Q183: How would you migrate a 100TB on‑prem Oracle data warehouse to the cloud? Expert
"Step 1: assess and refactor PL/SQL. Step 2: choose cloud target (e.g., Snowflake). Step 3: initial bulk load via Snowball or direct connect. Step 4: set up CDC for incremental sync until cutover. I led a migration where we used AWS DMS and cut over with zero downtime – users saw no difference."
Q184: How do you handle real‑time data in a data warehouse? Expert
"Combine a streaming layer (Kafka → Flink) with micro‑batch loading into the DW every 1‑5 minutes. Use merge statements for upserts. I built a real‑time inventory dashboard that updated every 2 minutes using Snowflake Streams and Tasks."
Q185: What is a data mesh? How does it change the data warehouse paradigm? Expert
"Data mesh decentralises ownership – each domain owns its data products, which are discoverable and interoperable. The traditional monolithic warehouse becomes a mesh of interconnected domains. I'm helping a retailer transition to mesh: each business unit publishes data as products via a shared governance layer."
Q186: What is query federation? When would you use it? Expert
"Querying data across multiple sources without moving it (Trino, Presto). I use it for ad‑hoc exploration across a lake and a warehouse, then materialize the useful join into a table."
Q187: How do you optimise a data warehouse for AI/ML workloads? Expert
"Provide a feature store that aggregates from the warehouse, store historical snapshots for training, and offer in‑database ML (BigQuery ML, Snowpark). I built a churn prediction pipeline where all features came from the DW, model trained weekly, and scores written back to the DW for BI."
Q188: Explain a data vault architecture with a concrete example. Expert
Hub_Customer (Customer_HK, Customer_BK, LoadDate) Link_Order_Customer (Order_Customer_HK, Order_HK, Customer_HK, LoadDate) Sat_Customer_Details (Customer_HK, Name, Address, LoadDate, RecordSource)
Q189: What is a "raw vault", "business vault", and "information mart" in Data Vault 2.0? Expert
"Raw Vault: Hubs, Links, Satellites untouched. Business Vault: derived, rule‑based transformations (effectivity satellites, PIT tables). Information Mart: star‑schema views for consumption."
Q190: What are "hash keys" in Data Vault? Why use them? Expert
"MD5/SHA hashes of business keys – they allow parallel loading and are independent of source sequence. I use hash keys to decouple ETL from source systems."
Q191: How to handle GDPR right‑to‑erasure in a data warehouse? Expert
"Anonymise or delete PII from all layers. I use a hashed mapping table; when a deletion request comes, I overwrite or tokenise the customer key and propagate to all tables via ETL that checks the deletion registry."
Q192: What is a data clean room? Expert
"A secure environment where multiple parties can analyse combined datasets without exposing raw data. I implemented one using Snowflake Secure Data Sharing with differential privacy."
Q193: What is horizontal vs vertical scaling in a data warehouse? Expert
"Horizontal: add more nodes (Redshift clusters, BigQuery slots). Vertical: increase single node power (Snowflake virtual warehouse size). I prefer horizontal for parallel workloads."
Q194: How do you implement a data warehouse on a Kubernetes cluster? Expert
"Use Trino + MinIO, or Apache Druid. I containerised our analytics stack and achieved elastic scaling for ETL tasks."
Q195: What is "query pruning" in the context of partitioned tables? Expert
"The optimizer skips partitions not needed based on filter. I design partitions around the most common WHERE clause (usually date) to drastically reduce scan size."
Q196: How to debug a hanging ETL job? Expert
"Check locks, skewed data distribution, resource contention. I use query profiler and monitoring (CloudWatch, Stackdriver). Once it was a missing commit that left a transaction open."
Q197: What is "workload management" (WLM) in a data warehouse? Expert
"Rules to prioritise queries – short BI queries vs long ETL. In Redshift, I configured manual WLM queues to prevent ETL from starving dashboard users."
Q198: How do you secure a cloud data warehouse? Expert
"Network isolation (private endpoints), encryption at rest and in transit, IAM roles, column‑level masking, and audit logging. I set up automated compliance checks."
Q199: What are the trade‑offs of using a columnar store vs row store for DW? Expert
"Columnar: better compression, high throughput for analytical queries. Row: better for point lookups and updates. DWs use columnar almost exclusively."
Q200: How to benchmark a data warehouse platform? Expert
"Use TPC‑DS or a custom workload reflecting your queries. I tested 10 typical queries with concurrent users, measuring latency and cost. That drove our cloud DW choice."
Q201–Q260: 60 Expert rapid‑fire Expert
Q201: What is a materialized view and how does it differ from a regular view? Stored result; faster but needs refresh.
Q202: Explain query rewrite. Oracle DB can automatically use materialized views.
Q203: What is a "pinned" dimension? A dimension that rarely changes – I store it in memory.
Q204: What is a "volatile" dimension? Changes frequently – handle with Type 1 or mini‑dimension.
Q205: What is a "dirty read"? Reading uncommitted data – in DW we avoid by using read‑committed isolation.
Q206: What is a "phantom read"? New rows appear in a repeated query – acceptable in most DW.
Q207: What is a "data skipping" index? Metadata that allows skipping blocks (e.g., Delta Lake file statistics).
Q208: How to implement a rolling window aggregate in SQL? Use window functions with ROWS BETWEEN.
Q209: What is a "conformed dimension" example? Date dimension – used by all facts.
Q210: What is the "bus architecture" in Kimball? A matrix linking business processes and dimensions.
Q211: How to design a DW for multi‑tenancy? Separate schemas, row‑level security, or dedicated compute pools.
Q212: What is a "hybrid architecture"? Combining on‑prem and cloud – I use a cloud DW for cold data and on‑prem for hot.
Q213: What is a "logical data warehouse"? A virtualisation layer over multiple physical stores.
Q214: What is "data fabric"? An architecture that automates integration and governance across hybrid landscapes.
Q215: How to handle schema‑on‑read? Use external tables with schema inference – good for exploration.
Q216: What is an "external table"? Data stored outside the DW (S3, ADLS) but queried as if inside.
Q217: What is "predicate pushdown"? Filter evaluation close to storage, minimising data transfer.
Q218: What is "projection pushdown"? Only reading needed columns in columnar stores.
Q219: What is a "query plan"? How the database executes a query – I always check it for optimisation.
Q220: What is a "hash join" vs "nested loop join"? Hash join builds hash table – better for large sets.
Q221: What is a "merge join"? Requires sorted inputs – fast but needs sort order.
Q222: What is "data skew"? Uneven distribution of data – I salt keys to fix it.
Q223: What is a "broadcast join"? Replicating a small table to all nodes – I use it in Spark.
Q224: What is "shuffle" in distributed processing? Redistribution of data across nodes – expensive, minimise it.
Q225: What is "cost‑based optimisation"? The DB chooses the cheapest plan based on statistics.
Q226: What are "statistics" in a DB? Metadata about data distribution – keep them up‑to‑date.
Q227: How to analyse a query plan? Look for full table scans, excessive shuffles, missing indexes.
Q228: What is a "materialized query table"? DB2 term for materialized view.
Q229: What is "dynamic partition pruning"? Optimisation where filters on dim propagate to fact partitions.
Q230: What is "zone map"? Min/max values per block – enables data skipping.
Q231: What is "bloom filter"? Probabilistic structure to test set membership – used for fast joins.
Q232: What is "vectorised execution"? Processing data in batches using SIMD – modern columnar engines use it.
Q233: What is "just‑in‑time compilation" (JIT)? Compiling query expressions to native code – I see it in Redshift.
Q234: What is "result set caching"? Reusing query results – BigQuery and Snowflake do it automatically.
Q235: What is "concurrency scaling"? Adding transient capacity for spikes – Snowflake’s multi‑cluster warehouses.
Q236: What is "auto‑suspend" in Snowflake? Stops compute when idle to save cost – I set it to 5 minutes.
Q237: What is a "data pipeline orchestration tool"? Airflow, Dagster – I schedule and monitor ETL.
Q238: What is "CI/CD for data pipelines"? Automated testing and deployment – I use dbt with GitHub Actions.
Q239: What is "data contract testing"? Validating that producers don't break consumer expectations.
Q240: What is a "schema registry"? Central repository of schemas (Avro/Protobuf) for data streams.
Q241: What is "DataOps"? Agile, DevOps for data – I promote automation and monitoring.
Q242: What is "observability in data pipelines"? Monitoring data freshness, volume, schema – I use Monte Carlo.
Q243: What is "reverse ETL"? Push data from DW back to operational tools (Salesforce, Braze).
Q244: What is a "customer data platform" (CDP)? Unified customer database – often built on DW.
Q245: What is "data democratisation"? Making data accessible to non‑technical users while governed.
Q246: What is a "semantic layer"? A business representation of data (LookML, dbt Metrics) – ensures consistent definitions.
Q247: What is "data literacy"? The ability to read, work with, and communicate data – I run training programs.
Q248: What is a "data product manager"? Manages data as a product, focusing on user needs.
Q249: How do you measure the success of a data warehouse? Adoption, query performance, data quality, and business impact.
Q250: What is "data downtime"? Periods when data is stale or inaccurate – I aim for <1h.
Q251: What is "data masking"? Hiding sensitive data (e.g., SSN) with realistic but fake values.
Q252: What is "tokenisation"? Replacing sensitive data with a token – used for PCI compliance.
Q253: What is "differential privacy"? Adding noise to query results to protect individuals – I use it for public dashboards.
Q254: How to implement row‑level security? Use secure views or policies – I map user attributes to data filters.
Q255: What is a "zero‑trust architecture" for data? Never trust, always verify – every access is authenticated and authorised.
Q256: What is "data classification"? Labelling data by sensitivity (public, internal, confidential).
Q257: What is a "data breach" response plan? We have an incident response runbook including communication and forensics.
Q258: What is "data ethics"? Ensuring data use respects privacy, fairness, and transparency.
Q259: How to balance data accessibility with security? Least‑privilege access with self‑service request workflows.
Q260: What is the role of a Chief Data Officer? Drive data strategy, governance, and data as an asset.
👑 Most Expert – AI‑Driven Warehouses & Future (60+ Q&As)
Q261: How would you build an AI‑powered autonomous data warehouse? Most Expert
"Embed ML for workload management (auto‑scaling), self‑tuning indexes, anomaly detection on data quality, and natural language querying. I architected a prototype using BigQuery ML for smart caching and a GPT‑4 interface that translates English to SQL – it reduced ad‑hoc requests by 70%."
Q262: What is a feature store, and how do you integrate it with a DW for real‑time ML? Most Expert
"A feature store (Tecton/Feast) serves both batch (from DW) and real‑time (from streams) features consistently. I designed a pipeline where DW nightly aggregates are pushed to the feature store, and online features from Kafka are merged at inference time – latency under 20ms."
Q263: Explain how you would use an LLM to automatically generate data warehouse documentation and lineage. Most Expert
"I feed the schema, ETL code, and query logs to a GPT‑4 model with a prompt to generate data dictionary, lineage graphs, and even business definitions. We fine‑tuned it on our naming conventions – it now produces 90% accurate docs that a human reviews."
Q264: What is "declarative data management" using AI? Most Expert
"You state what data you need, and the system auto‑generates pipelines, optimises storage, and handles failure. I'm experimenting with LLM agents that write dbt models from a YAML spec – it's 10x faster."
Q265: How to design a data warehouse that supports both batch and real‑time ML inferencing? Most Expert
"Use a lakehouse with Delta Lake, which handles batch and streaming via a unified table. ML models are registered in MLflow and called via UDFs. Real‑time features are served from Redis populated by Flink, and batch features join via the warehouse key."
Q266: What is the role of "data contracts" in a data mesh, and how do you enforce them with AI? Most Expert
"Data contracts define schema, semantics, and SLAs. I use an AI agent that monitors production data against the contract and alerts domain owners if anomalies or schema drift occur – it even suggests fixes."
Q267: How can you use generative AI to create synthetic data for warehouse testing? Most Expert
"I prompt an LLM with a table schema and business rules to generate realistic synthetic data, preserving referential integrity. This populates a test environment without exposing sensitive data – we built 100TB of test data in a day."
Q268–Q320: 53 most‑expert visionary gems Most Expert
Q268: What is the "data cloud" vision? Interconnected, governed data across all platforms.
Q269: How to implement a "data product" with a contract? Clear owner, published schema, SLA, and quality metrics.
Q270: What is a "composable data stack"? Pick best‑of‑breed tools, integrate via open standards.
Q271: How to migrate from a legacy EDW to a data mesh without disruption? Domain by domain, with a transition layer.
Q272: What is a "data reliability engineer"? A new role focusing on data uptime and quality.
Q273: How to implement data quality as code? dbt tests, Great Expectations suites in CI/CD.
Q274: What is "active metadata"? Metadata that drives automation (e.g., auto‑scaling based on query patterns).
Q275: Explain the concept of "entropy" in data and how to combat it. Data degrades over time; we need active curation.
Q276: How to build a "self‑healing data pipeline"? Auto‑retry, fallback sources, ML‑based anomaly correction.
Q277: What is "data‑aware orchestration"? Orchestration that understands data dependencies, not just tasks.
Q278: How to integrate data warehouse with a knowledge graph? Link entities across domains – I did this for pharma R&D.
Q279: What is "semantic search" on a data warehouse? Vector embeddings of column descriptions for natural language discovery.
Q280: How to apply reinforcement learning in query optimisation? An agent learns to pick the best join order – research phase.
Q281: What is "in‑database ML" and its limitations? Great for simple models, but not for deep learning – I use external services then.
Q282: How to handle GDPR while keeping analytical value? Use privacy‑preserving techniques like k‑anonymity, differential privacy.
Q283: What is "homomorphic encryption"? Process encrypted data without decrypting – still too slow for most DW.
Q284: What is the future of SQL? English as the new SQL – copilots will translate, but SQL remains the precise engine.
Q285: How to foster an "AI‑first" data culture? Provide AI‑powered tools, training, and celebrate data‑driven wins.
Q286: What is a "data flywheel"? More data → better models → better products → more users → more data.
Q287: How to measure ROI of a data warehouse modernisation? Track time‑to‑insight, cost per query, and new revenue enabled.
Q288: What is the "data economy"? Data as a tradable asset – we monetise aggregated insights.
Q289: How to build a data marketplace? Catalogue, pricing, and secure data sharing – Snowflake Marketplace is a start.
Q290: What is "data collaboration"? External partners sharing data in clean rooms for mutual benefit.
Q291: What is "edge data warehouse"? Lightweight DW on edge devices for real‑time local analytics – emerging.
Q292: How to use digital twins with a data warehouse? Integrate IoT streams and simulation models into the DW for what‑if analysis.
Q293: What is "observability‑driven development" for data? You instrument pipelines first, then build.
Q294: How to lead a data warehouse transformation? Start with a vision, get executive sponsorship, deliver quick wins, iterate.
Q295: What is the "Unified Analytics" platform? Databricks Lakehouse – one platform for DW, ML, streaming.
Q296: How to align data strategy with ESG goals? Measure carbon footprint of data storage and compute, optimise.
Q297: What is a "serverless data warehouse"? No provisioned resources – pay per query (BigQuery, Athena).
Q298: What is "multi‑cloud data warehouse"? A DW that runs across AWS, Azure, GCP – still aspirational.
Q299: How to ensure business continuity for a global DW? Multi‑region replication, disaster recovery exercises.
Q300: What is the "data mesh" maturity model? Explorer → Adopter → Scaler → Native – I assess organisations.
Q301: What is "data product thinking"? Treating data like a product with user research and roadmap.
Q302: How to use Jupyter notebooks with a DW? Via SQL magic – we do exploratory analysis directly on Snowflake.
Q303: What is a "lakehouse" for machine learning? Delta Lake + MLflow + AutoML – unified experience.
Q304: What is the difference between a data warehouse and a feature store? DW stores historical data; feature store serves pre‑computed ML features.
Q305: How to scale data validation at a petabyte scale? Automated profiling with Spark, rules as code.
Q306: What is "data anomaly detection" powered by AI? Unsupervised models that alert on unusual patterns – we use it for revenue monitoring.
Q307: How to use large language models for data cleaning? LLMs standardise addresses, categorise free text – I built a customer support ticket classifier using GPT‑4.
Q308: What is "text‑to‑SQL" and its current accuracy? LLMs translating English to SQL – now ~90% on standard schemas, but still needs human review.
Q309: How do you prevent prompt injection in an LLM‑powered analytics bot? Sanitise inputs, use a separate model for intent classification.
Q310: What is "AI TRiSM" (Trust, Risk, Security Management)? Framework to ensure AI models are fair, secure, and reliable – I apply it to ML in DW.
Q311: What is a "data lakehouse" vs a "data warehouse"? Lakehouse adds transactional support on lake storage – blurring the line.
Q312: How to do A/B testing with a data warehouse? Store experiment assignments in a dimension, use for analysis.
Q313: What is "decision intelligence"? Combining data, analytics, and AI to automate or augment decisions.
Q314: How to embed analytics into operational apps using the DW? Reverse ETL + embedded dashboards.
Q315: What is "streaming data warehouse"? Like Materialize – continuously updated SQL views on streams.
Q316: What is a "real‑time OLAP cube"? Apache Pinot, ClickHouse – sub‑second queries on fresh data.
Q317: How to choose between a specialised real‑time DB and a DW? If latency < 1s and high concurrency, use a real‑time DB; then later sync to DW for historical.
Q318: What is the "modern data stack" of 2026? dbt, Snowflake/Databricks, Airflow/Dagster, data observability, and AI copilots.
Q319: How to future‑proof a data warehouse? Adopt open formats (Iceberg), decouple compute/storage, and invest in data culture.
Q320: Final golden advice: "A data warehouse is a reflection of your business. If you model it with care, it will illuminate your path to growth."
🔬 Hands‑On Labs (20+ Code & SQL Exercises)
Lab 1: Create a star schema in SQL (PostgreSQL). Intermediate
CREATE TABLE dim_date (date_key INT PRIMARY KEY, full_date DATE, month INT, year INT);
CREATE TABLE fact_sales (date_key INT REFERENCES dim_date(date_key), product_id INT, amount DECIMAL);Lab 2: SCD Type 2 merge in BigQuery. Expert
MERGE dim_customer T
USING stg_customer S ON T.customer_bk = S.customer_bk AND T.is_current = TRUE
WHEN MATCHED AND (T.name <> S.name) THEN UPDATE SET is_current = FALSE, end_date = CURRENT_DATE()
WHEN NOT MATCHED THEN INSERT (customer_bk, name, start_date, is_current) VALUES (S.customer_bk, S.name, CURRENT_DATE(), TRUE);Lab 3: Redshift distribution and sort key creation. Expert
CREATE TABLE sales (
sale_id INT, date_id INT, product_id INT, amount DECIMAL
) DISTKEY(product_id) SORTKEY(date_id);Lab 4: Snowflake zero‑copy clone for dev environment. Expert
CREATE DATABASE DEV_DW CLONE PROD_DW;Lab 5: dbt model with incremental load. Intermediate
{{ config(materialized='incremental', unique_key='date_key') }}
SELECT date_key, SUM(amount) as total_sales
FROM raw_orders
{% if is_incremental() %} WHERE order_date > (select max(date_key) from {{ this }}) {% endif %}
GROUP BY 1Lab 6: Python script to generate date dimension. Intermediate
import pandas as pd
dates = pd.date_range('2020-01-01', '2030-12-31')
dim_date = pd.DataFrame({'date_key': dates.strftime('%Y%m%d').astype(int),
'full_date': dates, 'month': dates.month, 'year': dates.year})Lab 7: Airflow DAG to run dbt daily. Expert
from airflow import DAG
from airflow.operators.bash import BashOperator
with DAG('dbt_daily', schedule='@daily') as dag:
dbt_run = BashOperator(task_id='dbt_run', bash_command='dbt run')Lab 8: Create a BigQuery ML model for sales forecast. Expert
CREATE OR REPLACE MODEL `mydataset.sales_forecast`
OPTIONS(model_type='ARIMA', time_series_timestamp_col='date', time_series_data_col='sales') AS
SELECT date, sales FROM `mydataset.daily_sales`;Lab 9: Spark ETL to read Parquet and write to Delta Lake. Expert
df = spark.read.parquet("s3://raw/sales")
df.write.format("delta").mode("overwrite").save("/delta/sales")Lab 10: Query a Snowflake stream for CDC. Expert
CREATE STREAM my_stream ON TABLE my_table;
SELECT * FROM my_stream WHERE METADATA$ACTION = 'INSERT';Lab 11‑20: Additional labs Intermediate
Lab 11: Trino federated query joining PostgreSQL and MongoDB.
Lab 12: dbt snapshot for SCD Type 2.
Lab 13: Redshift UNLOAD to S3 with Parquet.
Lab 14: Snowflake data sharing setup.
Lab 15: Azure Synapse pipeline copy activity.
Lab 16: Implement row‑level security in BigQuery.
Lab 17: Great Expectations validation suite on a warehouse table.
Lab 18: Use Python to call OpenAI API to generate SQL.
Lab 19: Build a real‑time dashboard with Kafka + ClickHouse.
Lab 20: Automate data quality report using dbt tests and Slack alerts.
🎯 Real‑World Scenarios & Business Cases (15 In‑Depth)
Scenario 1: CEO wants a single view of customer across 10 source systems. Most Expert
"I'd use MDM to create a golden customer record, then build a Customer 360 dimension in the warehouse. We hash and match on email/phone, use survivorship rules for conflicting data. Delivered in phases – first marketing systems, then support. The unified view saved $2M in duplicate marketing."
Scenario 2: Your nightly ETL window is shrinking – what do you do? Expert
"Profile the bottleneck – usually a slow transformation. I'd push transformations to ELT, use incremental loads, and optimise SQL. Implemented streaming CDC and micro‑batching, cutting load time from 4 hours to 20 minutes."
Scenario 3: Data warehouse migration to cloud with zero downtime. Expert
"Keep on‑prem running, replicate data to cloud in near real‑time, run validation queries in parallel. Switch BI tools to cloud after thorough testing. We achieved 100% consistency and cut over during a maintenance window."
Scenario 4: Data model needs to support a new business line quickly. Intermediate
"I extend the existing bus matrix – add a new fact and link to conformed dimensions. Used an agile modelling workshop and had a working prototype in 3 days."
Scenario 5: A critical dashboard shows wrong numbers. Intermediate
"I trace lineage back to source, find a filter changed. Added automated data quality checks and alerts. Now we catch it before business sees it."
Scenario 6: How to design a data warehouse for a fintech that must comply with SOX. Expert
"I enable full audit trail: every change logged, SCD Type 2 for critical dimensions, and no direct deletes – only logical. Separate environments for dev/test. Implemented change control for all DW objects."
Scenario 7: You discover a 2% data discrepancy between source and warehouse. Expert
"Perform reconciliation: row counts, checksums, then drill down by partition. Found a timestamp rounding bug in the ETL. Fixed and re‑loaded affected partitions."
Scenario 8: Business users want self‑service analytics but data is complex. Expert
"I build a semantic layer with Looker/dbt Metrics, provide pre‑built data marts, and train power users. Adoption increased 60% after we launched a 'data champions' program."
Scenario 9: Manage a DW that ingests 1 billion events/day. Most Expert
"Use a streaming ingestion (Kinesis → S3), then micro‑batch into Snowflake with auto‑scaling. Partition by hour, cluster by user. I tune warehouse size dynamically based on load."
Scenario 10: AI model training needs daily snapshots from the warehouse. Expert
"I create a feature store that queries the warehouse daily, snapshots the data with a version, and serves it to the training pipeline. Ensures reproducibility."
Scenario 11: GDPR request forces you to delete a user’s data. Expert
"We use a deletion table – all ETL jobs check it and exclude/anonymise the user. Data is physically purged from all layers within 30 days."
Scenario 12: Your CEO asks, "Are we making money on this product line?" Beginner
"I pull the fact table for sales and costs joined with product dimension, group by product line, calculate margin. I present a trend over 12 months – turns out it’s declining, so we adjust pricing."
Scenario 13: Multiple teams keep creating their own siloed data marts. Expert
"I implement a governed self‑service model: centralised conformed dimensions in the EDW, teams create their own marts on approved infrastructure. Governance council reviews new marts."
Scenario 14: You need to reduce cloud data warehouse costs by 30%. Expert
"Analyse query patterns: suspend idle clusters, right‑size warehouses, implement data life‑cycle policies to archive old partitions to cheaper storage. Achieved 35% savings."
Scenario 15: How to build a data warehouse that supports AI copilot queries? Most Expert
"I create a rich metadata layer describing every table and column in natural language, plus a vector index of query logs. The copilot uses this to generate accurate SQL. We fine‑tuned on our internal queries."
About Md Mominul Islam
No comments:
Post a Comment
Thanks for your valuable comment...........
Md. Mominul Islam