Autocoder

Media.net Interview Questions and Answers

Last updated 2026-11-19

Media.net is an advertising technology company, so its engineering interviews commonly mix classic computer science fundamentals with ad-tech flavoured design: auctions, click-through rates, high-throughput counting and low-latency serving. This page offers practice questions on those themes, with worked answers.

They are practice questions in the style of the topics commonly reported, not real papers. For timed problems see Media.net coding round questions, and for design practice see system design interview questions.

Rounds as commonly reported

Candidate accounts describe an online test with coding and sometimes aptitude questions, followed by two or three technical rounds on data structures, algorithms and core subjects such as operating systems and databases, then a discussion with a manager or HR. Experienced hires add a design round. Media.net has not published a fixed process, so use this as an approximate pattern.

What to emphasise

Strong fundamentals matter: you may be asked why something works, not only how to use it. Be ready to justify data structure choices and to reason about throughput and latency numbers.

Hiring status

We hold no sourced hiring dates, numbers or process facts for Media.net and there is no /hiring page for it on this site, so none are stated. For current roles, use the company's own careers site.

Media.net Interview Questions and Answers: questions with worked answers

  1. 1

    How does a second-price auction work, and why is bidding your true value optimal?

    medium

    The highest bidder wins but pays the second-highest bid. If you bid your true value v, you win only when the price is at most v, so you never pay more than the item is worth to you. Bidding higher risks winning at a price above v; bidding lower risks losing an item you valued above the price. Bidding v is therefore a dominant strategy. Many ad exchanges have moved towards first-price auctions, where bidders shade their bids, so name both.

  2. 2

    Design a system that counts ad impressions and clicks at very high throughput.

    hard

    Receive events at stateless collectors, write them to a partitioned log such as Kafka keyed by campaign id, and aggregate in a stream processor into time windows (per minute, per campaign). Store the aggregates in a time-series or columnar store for dashboards, and keep raw events in cheap storage for reprocessing. Handle duplicates with event ids and idempotent writes, and late events with watermarks. Accept eventual consistency for reporting but not for billing, which should reconcile from raw logs.

  3. 3

    Two ads: A has 1 click in 2 impressions, B has 90 clicks in 100. Which has the better CTR?

    medium

    The raw rates are 50 percent and 90 percent, but A has almost no evidence. Apply a prior, such as adding a pseudo-count: with 1 pseudo-click in 10 pseudo-impressions, A becomes (1+1)/(2+10), about 17 percent, and B (90+1)/(100+10), about 83 percent. Alternatively, compare lower confidence bounds, such as the Wilson interval. This is the same intuition behind exploration in bandit algorithms.

  4. 4

    Explain virtual memory and paging.

    medium

    Each process sees a private virtual address space divided into pages, mapped by page tables to physical frames; the TLB caches recent translations. If a page is not present, a page fault makes the OS load it from disk, evicting another page by a policy such as LRU approximations. This gives isolation, lets programs exceed physical memory, and simplifies allocation. Thrashing occurs when the working set does not fit and the system spends its time paging.

  5. 5

    Compare FCFS, SJF, round robin and priority scheduling.

    easy

    FCFS is simple but suffers the convoy effect. Shortest job first minimises average waiting time but needs burst estimates and can starve long jobs. Round robin gives each process a time quantum, good response time, with context switch overhead if the quantum is too small. Priority scheduling serves important work first and needs ageing to prevent starvation. Real systems use multilevel feedback queues.

  6. 6

    Write SQL to find campaigns with CTR above 2 percent and at least 1000 impressions.

    medium

    Aggregate impressions and clicks per campaign, filter with HAVING, and compute CTR as clicks divided by impressions, using a numeric cast to avoid integer division. The minimum-impressions filter avoids picking campaigns on tiny samples.

    sql
    SELECT campaign_id,
           SUM(clicks) * 1.0 / SUM(impressions) AS ctr
    FROM daily_stats
    GROUP BY campaign_id
    HAVING SUM(impressions) >= 1000
       AND SUM(clicks) * 1.0 / SUM(impressions) > 0.02;
  7. 7

    What is a Bloom filter and where would an ad server use one?

    medium

    A Bloom filter is a bit array with k hash functions: adding sets k bits, and a lookup says "possibly present" if all are set or "definitely absent" otherwise. It has no false negatives but a tunable false positive rate and uses little memory. An ad server could use it to cheaply skip users who have already seen a capped ad, or to dedupe event ids, accepting occasional false positives.

  8. 8

    Explain consistent hashing.

    medium

    Map both servers and keys onto a hash ring; each key belongs to the next server clockwise. Adding or removing a server moves only the keys in its neighbouring arc, roughly 1/n of the data, instead of reshuffling almost everything as modulo hashing does. Virtual nodes (many points per server) even out load and let you weight bigger machines.

  9. 9

    Design an in-memory key-value store with expiry (TTL).

    medium

    Keep a hash map from key to value and expiry time. On get, check the expiry and delete lazily if past. Run a background sweep that samples keys or uses a min-heap of expiry times to remove expired entries so memory is reclaimed. Guard with a read-write lock or shard the map into segments to reduce contention.

  10. 10

    Reverse a linked list in groups of k.

    hard

    Count k nodes ahead; if fewer than k remain, leave them. Otherwise reverse those k nodes in place, attach the reversed group to the previous tail, and continue. O(n) time, O(1) space. The usual mistake is losing the pointer to the node after the group before reversing.

    python
    def reverse_k(head, k):
        probe, n = head, 0
        while probe and n < k:
            probe, n = probe.next, n + 1
        if n < k:
            return head
        prev, cur = None, head
        for _ in range(k):
            cur.next, prev, cur = prev, cur, cur.next
        head.next = reverse_k(cur, k)
        return prev
  11. 11

    Why do you want to join Media.net?

    easy

    Link it to the work: large-scale, low-latency systems and machine learning applied to real revenue. Name one technical challenge, for example real-time bidding under a tight time budget, and what you would like to build. Show that you understand the business in a sentence.

  12. 12

    Tell me about your biggest technical challenge.

    easy

    Pick a problem with a real obstacle, such as a memory leak or a slow query. Walk through how you diagnosed it (profiling, logs, hypothesis testing), what options you weighed, what you chose and the measured improvement. Be specific about your own part in a team effort.

Frequently asked questions

Does Media.net ask operating systems questions?

Core CS subjects such as operating systems and DBMS are commonly reported. Revise memory management, scheduling and synchronisation.

Is ad-tech knowledge required?

Not for most engineering roles, but knowing how auctions and CTR work helps in design discussions.

How difficult are the coding questions?

Commonly reported as medium to hard. See the Media.net coding round questions.

Where do I look for openings?

Use the Media.net careers site. This site has no sourced hiring data for the company.

Related guides

More in Company Interview Questions

See all in this section