Flipkart Interview Questions

OLTP (Online Transaction Processing): normalized schema, optimized for high volume inserts/updates/deletes, row oriented storage, low latency per transa...

• What is the difference between OLAP and OLTP systems?

  OLTP (Online Transaction Processing): normalized schema, optimized for high volume inserts/updates/deletes, row oriented storage, low latency per transaction queries. Examples: order management, payment systems. OLAP (Online Analytical Processing): denormalized star/snowflake schema, optimized for complex aggregations across millions of rows, column oriented storage, used for reporting and BI. In e commerce: OLTP handles live orders; ETL moves data nightly to OLAP for business intelligence.

• Describe your experience with data modeling. What types of data models have you worked with?

  Star schema : fact table (orders, events) + dimension tables (customer, product, date, seller). Most common for BI workloads — simple joins, fast aggregations. Snowflake schema : normalized dimensions for storage efficiency; more joins required. Data Vault : hub satellite model for auditability and historical tracking in enterprise DWHs. Wide/flat tables : pre joined denormalized tables for ML feature stores and direct dashboard consumption. One Big Table (OBT) : fully denormalized for maximum q

• How would you approach designing a data schema for an e-commerce platform?

  Fact tables : fact orders(order id, date key, customer key, seller key, product key, quantity, gmv, discount, delivery days) fact product views(view id, date key, customer key, product key, session id, source) fact returns(return id, order id, date key, reason key, refund amount) Dimension tables : dim customer(customer key, segment, city, tier, registration date) — SCD Type 2 dim product(product key, name, category, brand, seller id, price tier) dim seller(seller key, name, rating, fulfillment

• Can you explain the concept of data pipelines and their importance?

  A data pipeline is a series of automated steps that move and transform data from source systems to destinations. Stages: Extract (read from source) → Transform (clean, enrich, aggregate) → Load (write to target). Why important : Powers business intelligence and reporting Feeds ML models with training and feature data Enables real time decision making (fraud detection, recommendations) Decouples operational systems from analytical systems — no analytics load on production DBs At Flipkart scale: p

• What strategies do you use for debugging data processing jobs?

  1. Check logs first : Spark UI for failed stages/tasks; Airflow logs for task errors 2. Isolate the failure : run just the failing transformation on a small sample 3. Inspect input data : add show() and printSchema() before the failing step 4. Check data assumptions : nulls, unexpected values, schema mismatches at the boundary 5. Add row count checkpoints : log record counts at each transformation stage 6. Use .explain() : inspect Spark execution plan for unexpected full scans or missing pushdow

• How do you handle data deduplication in a data pipeline?

  At destination (Delta Lake) : Root cause fix: add unique constraint at source, use idempotent Kafka consumer offsets.

• How do you handle data versioning in a data pipeline?

  Delta Lake time travel : every write creates a new version. Query any past version: Partition based versioning : keep separate partitions for each processing run; don't overwrite dbt snapshots : SCD Type 2 for slowly changing dimensions — tracks row history with valid from / valid to Code versioning : all pipeline code in Git; tag releases. Each pipeline run logs the code version used. Metadata table : store run id , pipeline version , rows processed , source timestamp per execution

• Can you explain the concept of a data lake?

  A data lake is a centralized repository that stores all data — structured, semi structured, and unstructured — in its raw format until needed. Key characteristics : Schema on read : schema applied when querying, not when storing Any format : Parquet, JSON, CSV, images, logs, binaries Cheap storage : built on object storage (S3, ADLS, GCS) Flexible : supports ML, BI, streaming, and ad hoc analysis from the same store vs Data Warehouse : warehouses are schema on write, structured only, more expens