Interview guides · American Sugar Refining

American Sugar Refining Software Engineer Interview Guide

American Sugar Refining Software Engineer interview guide: reported rounds, coding, design and behavioral questions, and a 7-day prep plan.

The 60-second brief

American Sugar Refining is a sugar refining business, and candidate descriptions of its Software Engineer role centre on software for manufacturing processes and production facilities. Reported project types include automating processes, building new features for existing applications and integrating systems across different platforms. Candidates describe the engineer working alongside product management and operations to turn user needs into technical specifications, and also reviewing code, testing software and improving existing systems as business needs change.

This guide is for software engineers preparing for American Sugar Refining's Software Engineer loop. Candidates describe the role as needing a computer science background and roughly two to five years of experience, so the guide assumes you can already write production code in at least one language. It covers the three reported rounds, every question candidates report (debugging, the development lifecycle, languages, code quality, legacy performance, feature design, prioritisation, three coding questions, two design questions and three behavioral prompts) and a plan that turns them into practice sessions.

Candidates report three rounds spread over about 3-5 weeks: Initial Screenings, Technical Interviews and Behavioral Assessments. These stages are candidate descriptions, not a published process.

Focus areas

Software engineering fundamentals (role readiness)
Technical Q&A for job duties
Engineer panel interviews

Overview

Candidates describe the work as designing, building and maintaining applications that support manufacturing operations. Typical projects are automating a process, adding a feature to a system that already exists, and connecting systems that run on different platforms. The requirements candidates list name Java, Python or C#, SQL and NoSQL databases, Git, and Agile or Scrum practice, with AWS or Azure, CI/CD and machine learning as extras. That mix points at maintenance and integration work more than greenfield algorithm design, so expect to be asked how you find faults, change working systems safely and talk to non-engineers.

The reported questions are broad rather than language-specific. Technical and case-study prompts cover debugging a complex issue, the software development lifecycle, which languages you are most comfortable with, a past technical challenge, code quality and maintainability, a legacy system with performance problems, designing a feature for an existing application, and prioritising under tight deadlines with several stakeholders. The coding prompts are reversing a string with its time complexity, finding a shortest path in a graph, and comparing data structures. The design prompts are real-time inventory tracking across multiple locations and the considerations behind a scalable application.

Because the prompts are open, depth comes from concrete material you prepare in advance. Pick one language and be ready to write and analyse short code in it. Build four or five real stories (a hard bug, a performance fix, a quality practice you introduced, a technical argument you had to win, a conflict with a colleague) and decide which prompts each can serve. Work the design questions with a written component list and a stated consistency model, so the answer has structure instead of a list of buzzwords.

The American Sugar Refining interview process

Candidates describe three stages, from a screening through technical interviews to behavioral assessments. Each card below says what is reported and how to prepare for it.

Initial Screenings

Candidates describe Initial Screenings as the first stage, used to judge fit for the Software Engineer role. Treat it as the point where your background is matched against the requirements candidates list: a mainstream language such as Java, Python or C#, SQL and NoSQL experience, Git, and Agile or Scrum practice. The reported language question, asking which languages you are most comfortable with and why, is the kind of prompt to have a crisp answer ready for here.

What to demonstrate

Fit between your experience and the role as candidates describe it
How clearly you summarise your background and your language strengths
Whether your availability and expectations line up with the process

How to prepare

Write a short walk through your CV that names the language, database and version-control tools you have used in production, matching the stack candidates list.
Prepare the languages answer: one primary language with a specific example of why you chose it, one secondary language, and an honest limit of your experience in each.
Read the requirement list candidates describe and mark which items you can back with a project, so gaps are ones you can speak about calmly.

Technical Interviews

Candidates report Technical Interviews that test problem-solving skills and technical knowledge. The question types candidates report for this loop span debugging, the development lifecycle, code quality, a performance problem in a legacy system, feature design, coding (reversing a string, shortest path, data structure choice) and system design (inventory tracking, scalable architecture). Prepare each type separately: code should be written and traced by hand, and design answers should name components, data flow and failure behaviour.

What to demonstrate

Problem-solving skills and technical knowledge, as candidates describe the stage
Working code with correct complexity analysis for the coding questions
Structured reasoning when diagnosing a fault or designing a system

How to prepare

Write the string reversal in your main language two ways, an in-place two-pointer swap on a character array and a library or slicing call, and state time and extra space for each.
Implement breadth-first search for unweighted shortest paths and Dijkstra with a priority queue for weighted ones, and be ready to say which you chose and why.
Rehearse the legacy-performance answer as a sequence: reproduce, measure, find the bottleneck, change one thing, re-measure.
Sketch the inventory-tracking design on paper with an event stream, a per-location stock view and a reconciliation step.

Behavioral Assessments

Candidates describe Behavioral Assessments as the stage focused on teamwork, communication and alignment with company values. The behavioral prompts candidates report are about working with a difficult team member, the engineer's role in a cross-functional team, and advocating for a technical decision. Prepare stories where you personally made choices and can state the result, since the role is described as working with product management and operations.

What to demonstrate

Teamwork and communication, as the stage is described
How your examples line up with the company's stated values
Your own role and decisions in the stories you tell

How to prepare

Write one story each for the difficult team member, the technical advocacy and the cross-functional prompts, using situation, task, action and result, and cut each to what you personally did.
Read what the company publishes about its operations and values, then pick the one or two points that your examples honestly support.
Prepare a question for the interviewers about how engineers and operations staff work together day to day.

Where candidates lose points

Reversing a string without saying what the language allows

Strings are immutable in Java, Python and C#, so say whether you build a new string or swap in a character array, and state the extra space. Mention what happens with an empty string, one character, and text where one visible character spans several code units.

Answering the shortest-path question with an algorithm before asking about the graph

Ask first whether edges have weights, whether weights can be negative, and whether the graph is directed. Use breadth-first search for unweighted edges and Dijkstra when weights are non-negative, and state the complexity for each.

Proposing a rewrite for the legacy performance problem before measuring anything

Describe the order: reproduce the slowness, profile or read query plans and logs, locate the bottleneck, change one thing, and compare with the baseline. Mention a rollback plan, since the system is in daily use.

Designing real-time inventory as a single table updated in place

Define what real-time means, then separate the stream of receipts, shipments, transfers and adjustments from the current-stock view. Cover duplicate messages, out-of-order events, a location losing connectivity and reconciliation against physical counts.

Telling behavioral stories in terms of what the team did, or casting a colleague as the villain

Use first person singular for the decisions, name your own mistake or adjustment, and finish each story with a result you can state. Rehearse them aloud so each stays short enough to leave room for follow-ups.

Questions & practice

Write a function to reverse a string. What is the time complexity of your solution?

Medium · Coding / Algorithms

Approach

Clarify first: what counts as a character (bytes, UTF-16 code units, Unicode code points or user-visible graphemes), whether null or empty input is possible, and whether in-place modification is allowed.
Standard solution: copy into a character array and swap with two pointers moving inward from both ends until they meet. That is O(n) time and O(n) extra space, because strings in Java, C# and Python are immutable.
Name the library versions and their costs: s[::-1] in Python and new StringBuilder(s).reverse() in Java are both O(n); Java's StringBuilder.reverse keeps surrogate pairs together, while a naive char-array reverse in Java or C# splits them.
Call out the slow versions: building the result with repeated string concatenation in a loop copies the string each time and can be O(n^2); recursion is O(n) deep and fails on long inputs (Python's default recursion limit is 1000).
Test the edge cases: empty string, one character, even and odd lengths, a palindrome, and text with emoji or combining accents.

Follow-up

Reverse the order of the words in a sentence instead of the characters, in place.
How would you reverse a string correctly when it contains emoji and accented characters?
Can you reverse it with O(1) extra space? What has to be true about the input?

How would you approach solving a specific algorithm problem, such as finding the shortest path in a graph?

Medium · Coding / Algorithms

Approach

Ask about the graph before choosing: directed or undirected, weighted or not, negative weights possible, one source or all pairs, and its size.
Unweighted: breadth-first search from the source, O(V + E), because the first time BFS reaches a node is along a shortest path.
Non-negative weights: Dijkstra with a binary-heap priority queue, O((V + E) log V); skip popped entries whose distance is larger than the best already recorded, and keep a predecessor map to rebuild the path.
Negative weights: Bellman-Ford, O(V * E), which also detects a negative cycle if a distance still improves on the V-th pass; on a DAG, relax edges in topological order in O(V + E). For all pairs on a small dense graph, Floyd-Warshall is O(V^3).
Edge cases: unreachable target (return a sentinel such as -1 or infinity), source equals target, parallel edges, and grids where each cell is a node with up to four neighbours.

Follow-up

Why does Dijkstra give wrong answers with negative edge weights?
Find the cheapest path that uses at most k edges.
How would A* speed this up on a map, and what must be true of its heuristic?

Can you explain the differences between various data structures and when to use them?

Medium · Coding / Algorithms

Approach

Arrays and dynamic arrays (ArrayList, List<T>, Python list): O(1) index access, amortized O(1) append, O(n) insert or delete in the middle, and good cache locality. The default choice for sequential data.
Hash maps and sets (HashMap, Dictionary<TKey,TValue>, dict): average O(1) lookup, insert and delete, but no ordering and a worse case when many keys collide. Use them for membership checks, counting and caching.
Balanced search trees (TreeMap in Java, SortedDictionary in C#): O(log n) operations with keys kept sorted, so they support range queries and nearest-key lookups that a hash map cannot.
Heaps (PriorityQueue in Java and .NET, heapq in Python): O(log n) push and pop, O(1) peek at the minimum; use them for top-k, scheduling and Dijkstra. Stacks and queues (ArrayDeque, collections.deque) give O(1) operations at the ends for DFS, BFS and undo history.
Linked lists give O(1) insert or remove at a node you already hold but O(n) access by position and poor cache behaviour. Tries suit prefix search, and graphs use adjacency lists (O(V + E) memory) unless they are dense enough for a matrix (O(V^2)).

Follow-up

Design an LRU cache with O(1) get and put. Which structures do you combine?
When would a sorted array with binary search beat a hash map?
What happens to hash map performance when many keys hash to the same bucket?

Stop tag and share joins from fanning out a page

Medium · join fan-out · semi-joins · many-to-many

resource_tag is (resource_id, tag_id) with PK (resource_id, tag_id); resource_share is (resource_id, shared_with_user_id, permission). The tagged-and-shared listing inner-joins resource to both, filters tenant_id, tag_id = ANY($2) and shared_with_user_id = $3, orders by updated_at DESC and takes 50. Pages come back with fewer than 50 distinct resources and the total in the header is far too high. Explain the row multiplication, rewrite both the page query and the count query so each is correct, and name the index each one needs. PostgreSQL 16.

Approach

Do the arithmetic against the predicates that are actually there. An inner join emits one row per matching child row, and both joins are filtered: tag_id = ANY($2) admits only the requested tags, shared_with_user_id = $3 admits one user's share rows. So a resource holding three of the requested tags and shared with $3 once yields three rows, not one — the multiplier is its count of matching tags times its share rows for that single user, and that second factor is 1 unless the table admits duplicate (resource_id, shared_with_user_id) pairs. LIMIT 50 then limits rows rather than resources, and COUNT(*) counts pairs — the header is the product, not the population.
Reject DISTINCT as the fix. It deduplicates after the product has been built, so the planner must materialise and sort the fanned-out set before the LIMIT can apply, and it leaves any SUM or AVG in the same select list wrong.
Rewrite both filters as semi-joins, keeping resource as the only row source: AND EXISTS (SELECT 1 FROM resource_tag rt WHERE rt.resource_id = r.resource_id AND rt.tag_id = ANY($2)) and the same shape against resource_share. A semi-join stops at the first match per resource and preserves the driving index order, so ORDER BY updated_at DESC, resource_id DESC LIMIT 50 still stops after 50 rows.
Count with the same predicates and no join at all: SELECT count(*) FROM resource r WHERE r.tenant_id = $1 AND r.status = 'active' AND EXISTS (...) AND EXISTS (...). Nothing multiplies a resource, so the number is the population.

Follow-up

The filter changes from 'any of these tags' to 'all of these tags'. Rewrite it and state what it costs relative to the ANY form.
A resource can be shared with the same user twice under different permissions. Does your count change, and should it?

Denormalise tenant onto revisions and backfill it live

Hard · denormalisation · online migration · backfill

resource_revision (revision_id, resource_id, version, actor_user_id, change_kind, patch, request_id, created_at) has 400M rows and no tenant column; tenant_id lives only on resource. Two reads need it: a tenant-scoped audit feed ordered by created_at DESC, and an offboarding purge. Both join back to resource today. Justify adding tenant_id to resource_revision against those two reads, name the anomaly the copy introduces and the constraint that prevents it, then give the ordered migration for a live table taking 1.2k writes/second — the lock each step takes, how the backfill is batched, and where each step stops being reversible. PostgreSQL 16.

Approach

Justify from the access path rather than from taste. Without the column, the audit feed either scans resource_revision by created_at and discards other tenants' rows, or resolves the tenant's resource_ids first and probes with them — both proportional to the tenant's whole history rather than to one page. With (tenant_id, created_at DESC, revision_id DESC) it is a seek that stops at 50 rows, and the purge becomes a ranged delete instead of a join.
Name the cost exactly: a second copy of a fact can disagree with the first. Make the disagreement unwritable rather than documented — add UNIQUE (resource_id, tenant_id) on resource so it can serve as a foreign-key target, then FOREIGN KEY (resource_id, tenant_id) REFERENCES resource (resource_id, tenant_id) on the revision table. A revision can then only ever carry its parent's tenant.
Step one, expand: ALTER TABLE resource_revision ADD COLUMN tenant_id BIGINT NULL, with no default, so it is a catalogue change and no rewrite. It still needs ACCESS EXCLUSIVE for an instant, and that instant queues behind the longest open transaction on the table while every later query queues behind it — set lock_timeout to 2s and retry rather than wait.
Step two, dual-write: deploy the writer that populates tenant_id on every new revision while reads still use the join. Reversible by redeploying the previous build, because nothing reads the column yet.

Follow-up

The backfill is half finished and a rollback is required. What state is the table in, and what does the previous build do with a half-populated column?
How do you verify the backfill actually finished, given rows are still being inserted while it runs?

Can you explain the software development lifecycle and its importance?

Medium · Technical / Domain Questions

Approach

Name the phases in order: requirements, design, implementation, testing, deployment, maintenance, and say what each one produces (a requirement list, a design outline, code, test results, a release, fixes and changes).
Contrast models: waterfall runs the phases once in sequence; Agile methods such as Scrum, which the role requirements mention, repeat a short cycle every sprint so feedback arrives early.
Explain why it matters: a misunderstood requirement found in review costs a conversation, the same mistake found after release costs rework, a redeploy and sometimes data cleanup.
Tie it to practice: version control in Git, code review, automated tests and a CI/CD pipeline are how each phase gets checked without slowing the cycle down.
Give one example from your own work of a phase that was skipped or rushed and what it cost, in the first person.

Follow-up

When would you pick a more sequential model over Scrum?
Where does testing belong in an Agile sprint, and who owns it?
How does maintenance of a long-lived system change the way you design it?

What programming languages are you most comfortable with, and why?

Medium · Technical / Domain Questions

Approach

Lead with one primary language and give reasons tied to work you shipped, ideally one of the languages candidates report in the requirements for this role (Java, Python or C#).
Give technical reasons, not feelings: static typing and refactoring tools in Java or C#, fast iteration and a large data ecosystem in Python, the concurrency model, the build and test tooling you know well.
Name a real limitation of your favourite and how you work around it, for example that standard CPython threads do not run CPU-bound Python code in parallel, so CPU-heavy work goes to multiprocessing or native libraries.
Say how you choose a language for a new task: what the team already maintains, the libraries the problem needs, performance and deployment constraints.
Show you can move between them: one concrete case where you picked up a language or codebase you did not know and how long it took to be productive.

Follow-up

What do you dislike about the language you picked?
How does memory management work in that language, and when has it caused you a problem?
If the team works mostly in a language you know less well, how would you ramp up?

Describe a technical challenge you faced in a past project and how you overcame it.

Medium · Technical / Domain Questions

Approach

Pick a challenge with a technical cause you can explain end to end (a performance problem, a data inconsistency, a failing integration), not a schedule or people problem.
Set it up with numbers: what was slow or broken, how much, who was affected, and what constraint made the obvious fix unusable.
Spend most of the answer on the options you considered, why you rejected the others, and how you tested your choice before rolling it out; say "I" for your own decisions.
End with a measurable result (latency, error rate, hours saved) and what you would do differently now.
Avoid the story where the team solved it and your own part is unclear; that is the version that leaves nothing to assess.

Follow-up

What was the second-best option, and why did you reject it?
How did you confirm your change caused the improvement?
What would have happened if your fix had failed in production?

How do you ensure code quality and maintainability in your projects?

Medium · Technical / Domain Questions

Approach

Automate the checks that do not need judgment: a formatter and linter, static analysis (for example SpotBugs for Java, Roslyn analyzers for C#, ruff or mypy for Python) and the test suite, all run in CI on every pull request.
Use tests at the right level: fast unit tests for logic, a smaller set of integration tests for database and service boundaries, and a regression test for every bug fixed. Treat coverage as a signal, not a target.
Keep changes reviewable: small pull requests, branch protection in Git, and reviews that focus on design, naming, error handling and test gaps rather than style the tools already enforce.
Design for change: clear module boundaries, dependency injection where it makes code testable, no duplicated business rules, and short notes recording why a non-obvious decision was made.
Pay down debt deliberately: track it, refactor behind passing tests, and leave code a little cleaner each time you touch it.

Follow-up

How do you add tests to code that has none and was not written to be testable?
What do you do when a reviewer and you disagree on a design choice?
Which quality metric would you track for a team, and how could it be gamed?

Imagine you are given a legacy system that is critical for operations but is exhibiting performance issues. Ho

Medium · Problem-Solving / Case Studies

Imagine you are given a legacy system that is critical for operations but is exhibiting performance issues. How would you approach diagnosing and solving the problem?

Approach

Define the problem before touching code: which operations are slow, since when, measured how (p50, p95 and p99 latency, throughput, error rate), and whether it lines up with a deploy, data growth or a load change.
Measure where the time goes: a profiler or APM traces on the application, the database slow-query log and EXPLAIN plans on the worst queries, and host metrics for CPU, memory, garbage-collection pauses, disk I/O and lock waits.
Check the usual legacy culprits: missing indexes on tables that have grown, N+1 query patterns, connection-pool exhaustion, synchronous calls to slow dependencies, and unbounded caches or logs.
Change safely because the system is in daily use: write characterization tests that pin current behaviour, change one thing at a time, keep a rollback plan or feature flag, and compare against the baseline after each step.
Separate the quick fix from the long-term plan: an index or query rewrite now, and for deeper problems an incremental replacement (strangler pattern) instead of a big-bang rewrite.

Follow-up

There are no tests and the original authors are gone. Where do you start?
You cannot reproduce the slowdown outside production. What do you do?
The fix needs a schema change on a large table that cannot go offline. How do you roll it out?

Discuss the steps you would take to design a new feature for an existing application.

Medium · Problem-Solving / Case Studies

Approach

Start with the users and the problem: who needs the feature, what they do today, and acceptance criteria that say when it is done; confirm scope before designing.
Study the existing application: its data model, the modules the feature touches, existing patterns to follow, and what could break for current users.
Design the change: API and data-model additions, backward compatibility, and for schema changes an expand-and-contract migration (add the new column, backfill, switch reads and writes, then remove the old one).
Plan delivery in small increments behind a feature flag, with unit and integration tests, monitoring for the new path, and a rollback plan.
Close the loop: decide up front how you will measure whether the feature worked, and check it after release.

Follow-up

How do you handle a feature that needs a breaking change to an API other systems call?
What would you cut if the deadline moved up by half?
How would you know a month later whether the feature was worth building?

How would you prioritize tasks in a project with tight deadlines and multiple stakeholders?

Medium · Problem-Solving / Case Studies

Approach

List every task with its deadline, its owner on the stakeholder side and its dependencies, then find the critical path: the chain of tasks that sets the earliest finish date.
Rank by impact and cost of delay against effort; a simple split into must-have, should-have and could-have makes the trade-off visible to everyone.
Make conflicts explicit: when two stakeholders both need their item first, put the trade-off in writing and get the decision from whoever owns the priority, instead of quietly picking one.
Protect quality by cutting scope, not testing; ship the must-haves complete and move the rest to a dated follow-up.
Re-check priorities as new information arrives and tell stakeholders early when a date is at risk, with options rather than just bad news.

Follow-up

Two senior stakeholders give you conflicting priorities. What do you do?
How do you tell a stakeholder their request will not make the deadline?
What would make you push back on the deadline instead of cutting scope?

How would you design a system to track inventory levels in real-time across multiple locations?

Medium · System Design / Architecture

Approach

Pin down the requirements: which events change stock (receipt, transfer, consumption, shipment, count adjustment), how fresh dashboards must be, whether stock may ever go negative, and what happens when a site loses its network connection.
Make an append-only ledger of inventory movements the source of truth, each event carrying item, location, quantity, unit and an idempotency key so a retried message is not counted twice; current balances per item and location are a projection built from it.
Stream events through a log such as Kafka partitioned by item and location, which keeps per-key ordering; consumers update the balance table and push changes to dashboards over WebSockets or server-sent events, with a cache for the cross-site totals.
Choose consistency per operation: a reservation or withdrawal at one location needs a strong check, for example an UPDATE that subtracts the quantity only WHERE qty >= the requested bind parameter, while totals across locations can be eventually consistent.
Handle the hard cases: transfers as two events with an in-transit state, sites that buffer events locally while offline and sync later, unit-of-measure conversion, and periodic physical counts posted as adjustment events so the ledger stays auditable.

Follow-up

A site is offline for several hours and keeps recording movements. How do you reconcile when it reconnects?
Two orders try to take the last units at the same location at once. What prevents overselling?
A balance is found to be wrong. How do you trace and correct it?

Discuss the architectural considerations you would take into account when building a scalable application.

Medium · System Design / Architecture

Approach

Start from load and requirements: expected users and requests per second, the read/write ratio, latency targets, data growth and availability needs, so each later choice has a reason.
Keep the application tier stateless behind a load balancer so it scales horizontally; move session state to a shared store and use autoscaling on a cloud platform such as AWS or Azure, which candidates list as a nice-to-have for this role.
Scale the data layer in steps: indexes and query tuning first, then read replicas (accepting replication lag, which breaks read-your-own-writes unless those reads go to the primary), then caching with cache-aside and TTLs, and partitioning only when one primary can no longer take the writes.
Move slow or bursty work off the request path onto a queue, and design for failure: timeouts on every remote call, retries with exponential backoff and jitter, idempotent handlers, and circuit breakers.
Add observability (metrics, structured logs, traces) and CI/CD so changes ship safely, and say when a modular monolith is the better trade-off than microservices.

Follow-up

Traffic grows tenfold. What breaks first, and how would you find out?
How do you prevent a cache stampede when a popular key expires?
A user updates a record and immediately sees the old value. What happened and how do you fix it?

How do you approach debugging a complex issue in your code?

Medium · Technical / Domain Questions

Approach

Reproduce it first, then shrink it: find the smallest input, configuration or sequence of steps that still fails, because every later step runs faster on a small reproduction.
Read the whole error and stack trace, then check what changed: recent commits, dependency versions, config, data. If a known-good commit exists, git bisect binary-searches the history and finds the breaking commit in about log2(n) test runs.
Form one hypothesis at a time and test it with a breakpoint, an assertion or a targeted log line that would prove it wrong, rather than changing code and hoping.
Treat intermittent failures differently: suspect races, shared state, timeouts and environment differences, and add logging with timestamps and request or thread ids, since a debugger can hide timing bugs.
Fix the root cause, not the symptom, and add a regression test that failed before the fix and passes after it.

Follow-up

The bug only shows up in production and you cannot attach a debugger. What do you do?
How would you debug a failure that happens once in every few hundred runs?
How do you know the fix addressed the cause and did not just hide it?

Suggested practice checklist

The plan follows the loop as candidates report it: a screening, technical interviews covering debugging, coding and design, then behavioral assessments. Each day names the reported questions it practises and ends with something you can keep.

Day 1 — Map the loop and fix your language answer

List the three reported rounds and write beside each the question types you expect to prepare for.
Mark each item in the requirement list candidates describe (Java, Python or C#, SQL and NoSQL, Git, Agile or Scrum) with the project that proves it.
Write and say aloud your answer on which languages you are most comfortable with and why.

Deliverable: A one-page round map, a requirements-to-projects table, and a written languages answer.

Day 2 — Strings and data structures

Write string reversal in your main language as an in-place swap and as a library call, with time and space stated, plus test cases for empty, single-character and multi-code-unit text.
Make a comparison table of array, linked list, hash map, set, stack, queue, heap and balanced tree: lookup, insert and delete costs and when you would pick each.
Explain the table aloud as an answer to the data structures question.

Deliverable: A working reversal function with tests, and a one-page data structure comparison table.

Day 3 — Shortest path

Implement breadth-first search over an adjacency list and return the shortest path, not just the length.
Implement Dijkstra with a priority queue and write its complexity.
Write three sentences on when each algorithm fails: negative weights, disconnected nodes, cycles.

Deliverable: Two tested shortest-path implementations and a note on when to choose each.

Day 4 — Debugging and legacy performance

Write a debugging checklist from a real bug: reproduce, narrow the scope, form a hypothesis, test it, confirm the root cause, add a regression test.
Outline an answer for the legacy system with performance issues: measurement, bottleneck candidates (queries, I/O, memory, locks) and how you would roll changes out safely.

Deliverable: A debugging checklist, a written legacy-performance plan and two worked SQL exercises.

Day 5 — System design: inventory and scalable applications

Design real-time inventory tracking across locations on paper: events, per-location stock view, idempotent updates and reconciliation.
Write a short list of architectural considerations for a scalable application: stateless services, caching, database scaling, queues and monitoring.
Walk through how to add a new feature to an existing application, from requirements through rollout.

Deliverable: One inventory design diagram, a scalability checklist and a feature-delivery outline.

Day 6 — Engineering practice and delivery

Write a clear explanation of the software development lifecycle with an example of how each stage appeared in a project you worked on.
List the code quality practices you use (code review, automated tests, CI/CD, naming and structure) and one example where each helped.
Prepare the past technical challenge story, then answer the prioritisation question with a method for ranking tasks and talking to stakeholders.

Deliverable: Written answers for the lifecycle, code quality, technical challenge and prioritisation questions.

Day 7 — Behavioral rehearsal

Write situation-task-action-result notes for the difficult team member and technical advocacy prompts.
Write your position on a software engineer's role in a cross-functional team, with a supporting example.
Run all three answers aloud with a friend or a recording and trim until each can be followed in a single pass.

Deliverable: Three finished behavioral answers and a recording of at least one.

Behavioural preparation

The behavioral prompts candidates report for this loop cover conflict with a colleague, the engineer's role in a cross-functional team, and defending a technical decision. The role is described as working with product management and operations, so stories about translating needs, explaining trade-offs and handling disagreement are the most relevant.

Describe a technical challenge you faced in a past project and how you overcame it.
Tell me about a time you had to prioritise work with several stakeholders and a tight deadline.

Describe a situation where you had to work with a difficult team member. How did you handle it?

Medium · Behavioral / Leadership

Approach

Choose a conflict where the friction was observable and specific, such as repeated missed handoffs, dismissive code review comments or decisions made without telling you. State it neutrally in two sentences and avoid labelling the person.
Show your first move was a private, direct conversation aimed at understanding their constraints, followed by a concrete working agreement, for example a review turnaround, a written decision log or a short sync.
Own your part: name one thing you changed in your own behaviour. A story with no self-correction reads as blame.
Close with an observable result (reviews unblocked, fewer reverted merges, a launch that shipped) and say whether the working relationship improved. If it did not, say what you did to keep the work moving.

Follow-up

What would you have done if the direct conversation had changed nothing?
What did you learn about your own contribution to the friction?
How do you keep the project moving while a disagreement is unresolved?

What do you believe is the role of a software engineer in a cross-functional team?

Medium · Behavioral / Leadership

Approach

Answer with a position, then prove it with an example. A strong position is that the engineer turns what product and operations need into a buildable specification, says early what is costly or risky, and owns the quality of what ships.
List the concrete duties: asking operations users how the work happens today before designing, writing down trade-offs so non-engineers can decide, reviewing and testing code, and supporting the feature after release.
Tie it to the role described for this company: software that supports manufacturing and production facilities means your users are often operations staff, so describe how you would learn their workflow rather than assume it.
Back the position with one real example where you clarified an ambiguous request or pushed back on scope, and give what changed as a result. Avoid a generic answer such as 'I write the code'.

Follow-up

Product asks for a date you believe is unrealistic. What do you do?
How do you explain a technical constraint to someone who does not write code?
When operations reports a defect in production, who owns the fix and the follow-up?

Tell us about a time when you had to advocate for a technical decision. What was the outcome?

Medium · Behavioral / Leadership

Approach

Pick a decision with real alternatives, such as a migration approach, a library or datastore choice, a refactor, or adding automated tests to a fragile module. Name the options and the criteria you used to compare them.
Show the evidence you brought: a prototype, a benchmark, incident history or a cost estimate. A decision backed by data beats one backed by seniority or taste.
Identify who had to agree and what each person was worried about (delivery date, operational risk, maintenance cost), then describe how you adjusted the plan to address those concerns.
Report the outcome honestly, including partial adoption or being overruled, and what you did next. State a measurable result if there was one, and say what you would do differently.

Follow-up

What would you have done if the decision had gone the other way?
What did your approach cost, and who bore that cost?
How would you have known early if you were wrong?