Cognizant Data Engineer Interview Questions (2026)

1. 1GenC Online Assessment60 minMedium

   Aptitude, verbal, and basic technical MCQs; higher tiers (Elevate/Next) add progressively harder coding sections.

2. 2GenC Next Coding Assessment60 minHard

   2-3 coding problems plus SQL/full-stack questions that gate the top fresher package.

3. 3GenC Technical + HR Interview45 minMedium

   Combined panel covers projects, programming fundamentals, and HR questions on flexibility and joining in one sitting.

4. 4Technical Interview (Lateral)45 minMedium

   Skill/vertical depth: Java/.NET/cloud scenarios, or domain platforms like Facets/Guidewire for healthcare and insurance roles.

5. 5Managerial Round40 minMedium

   Delivery manager tests estimation, client escalation handling, and team leadership for experienced hires.

6. 6HR Discussion30 minEasy

   Compensation, location/shift preferences, notice-period negotiation, and documentation.

1. Q1

   Design the incremental pull for Cognizant's insurance client's claims data lake. Which watermark column do you trust, and what happens when the source clock drifts?

   HardSQL roundincremental extractionCognizant-specific

   Context: Cognizant insurance client's claims data lake

   How to answer: Pick an audit column the source actually maintains, overlap the window, and dedupe on a business key rather than assuming exactly-once. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

2. Q2

   Rows in Cognizant's e-commerce client's clickstream ingestion arrive more than once after a retry. Write the SQL that lands exactly one row per business key and say which one survives.

   MediumSQL rounddeduplication on loadCognizant-specific

   Context: Cognizant e-commerce client's clickstream ingestion

   How to answer: ROW_NUMBER over the business key ordered by load or update time, with an explicit survivorship rule you can defend to the client. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

3. Q3

   The client wants history preserved on the customer dimension behind Cognizant's healthcare client's patient-records batch feed. Write the SCD Type 2 merge and name what it costs.

   HardSQL roundslowly changing dimensionsCognizant-specific

   Context: Cognizant healthcare client's patient-records batch feed

   How to answer: Effective-from/to plus a current flag, hash-diff on tracked columns, and the row growth and query complexity you accept in return. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

4. Q4

   After a new join, the fact row count in Cognizant's logistics client's shipment-tracking pipeline tripled. Debug it and prove the fix.

   HardSQL roundjoin fan-out debuggingCognizant-specific

   Context: Cognizant logistics client's shipment-tracking pipeline

   How to answer: Find the one-to-many side, check key uniqueness first, pre-aggregate or dedupe, then verify with counts before and after. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

5. Q5

   A dimension key is missing when the fact for Cognizant's manufacturing client's sensor telemetry feed loads. What do you write instead of dropping the row?

   MediumSQL roundNULL and late-arriving keysCognizant-specific

   Context: Cognizant manufacturing client's sensor telemetry feed

   How to answer: Route to an unknown-member key, keep the fact, and reprocess when the dimension lands — never silently lose revenue rows. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

6. Q6

   The reconciliation query over Cognizant's media client's subscriber-events pipeline scans the whole table every night. How do you make it cheap?

   MediumSQL roundquery performanceCognizant-specific

   Context: Cognizant media client's subscriber-events pipeline

   How to answer: Partition pruning, predicate pushdown, and clustering on the filter column; read the plan before changing anything. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

7. Q7

   One Spark task in Cognizant's utilities client's smart-meter reading feed runs for an hour while the rest finish in a minute. Diagnose and fix it.

   HardPython and Spark rounddata skewCognizant-specific

   Context: Cognizant utilities client's smart-meter reading feed

   How to answer: Inspect partition sizes for a hot key, then salt, broadcast the small side, or repartition — and say which one you would try first and why. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

8. Q8

   Walk through what actually happens when Cognizant's travel client's booking and cancellation mart does a groupBy on a billion rows. Where does the time go?

   HardPython and Spark roundshuffle and partitioningCognizant-specific

   Context: Cognizant travel client's booking and cancellation mart

   How to answer: Wide vs narrow transformations, shuffle write and read, partition count, and spill to disk. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

9. Q9

   When would you broadcast the dimension in Cognizant's banking client's nightly core-banking ingestion, and when does that blow up?

   MediumPython and Spark roundbroadcast joinsCognizant-specific

   Context: Cognizant banking client's nightly core-banking ingestion

   How to answer: Small-side size against executor memory, the auto-broadcast threshold, and the driver OOM you get when you force it on a large table. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

10. Q10

    The landing zone for Cognizant's retail client's daily sales fact load has hundreds of thousands of tiny files. What is the damage, and what is the fix?

    HardPython and Spark roundsmall files problemCognizant-specific

    Context: Cognizant retail client's daily sales fact load

    How to answer: Listing and task overhead, then compaction, target file sizes, and fixing the writer rather than compacting forever. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

11. Q11

    A teammate solved a transformation in Cognizant's telecom client's call-detail-record pipeline with a Python UDF. When is that the wrong call?

    MediumPython and Spark roundUDFs vs built-insCognizant-specific

    Context: Cognizant telecom client's call-detail-record pipeline

    How to answer: Serialization cost and the lost optimizer pushdown; reach for built-in functions first, then pandas UDFs if you must. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

12. Q12

    The same intermediate DataFrame in Cognizant's insurance client's claims data lake is used four times. Do you cache it? Defend the answer.

    MediumPython and Spark roundcaching and reuseCognizant-specific

    Context: Cognizant insurance client's claims data lake

    How to answer: Recompute cost against memory pressure and eviction; measure with the DAG rather than caching by reflex. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

13. Q13

    State the grain of the fact table behind Cognizant's e-commerce client's clickstream ingestion in one sentence, and show what breaks when the grain is wrong.

    MediumData modelling roundgrain definitionCognizant-specific

    Context: Cognizant e-commerce client's clickstream ingestion

    How to answer: One row equals one what; a mixed grain gives double counting no downstream fix can rescue. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

14. Q14

    Model Cognizant's healthcare client's patient-records batch feed as a star schema. Which tables are facts, which are dimensions, and why not one wide table?

    MediumData modelling roundstar schema designCognizant-specific

    Context: Cognizant healthcare client's patient-records batch feed

    How to answer: Conformed dimensions, additive measures, and the update and consistency cost of a single flattened table. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

15. Q15

    Explain the bronze, silver and gold layers for Cognizant's logistics client's shipment-tracking pipeline, and what is not allowed to happen in each.

    MediumData modelling roundlayered architectureCognizant-specific

    Context: Cognizant logistics client's shipment-tracking pipeline

    How to answer: Raw fidelity in bronze, conformed and deduplicated in silver, business-shaped aggregates in gold; no business logic hidden in ingestion. State the requirement, the data you would move, and the checks that make the output trustworthy. Walk the design concretely: sources, storage layout, transformations, schedule, and the failure cases. Close with how you would validate the result and hand it over to whoever runs it next.

Practice these with instant AI feedback in a live mock interview → Start a Cognizant Data Engineer mock

Clear the aptitude and coding test; revise programming and SQL; prepare project and HR questions

Indicative Data Engineer pay in India: ~₹10–45 LPA (role-level range, not a Cognizant-specific figure).

Other roles at Cognizant

Data Engineer interviews at other companies