AST Software Engineer Interview Guide

The 60-second brief

Candidates describe AST as an industrial automation company serving the life science industry. Its software controls equipment used in aseptic manufacturing. According to the responsibilities candidates report, a Software Engineer here works in an R&D group and owns PC- and Android-based control software for automation machines. The work spans .NET applications, integrating machine learning models into that control software, working with SQL Server or PostgreSQL, unit and integration testing, and documentation, including operation manuals for the machines. Candidates also describe working across engineering teams to scope features and design their architecture before building them.

This guide is for engineers preparing for AST's Software Engineer loop. The reported requirements are production experience in C# and .NET Core/.NET 6+, SQL Server or PostgreSQL, end-to-end machine learning pipelines and Scrum/Agile. Regulated industrial or PLC work, desktop UI frameworks such as Avalonia, WPF or WinUI, and Unity are listed as nice to have. The guide covers the three reported stages, a full answer outline for each reported question, the database and ML scenarios candidates describe, and a seven-day plan built around them.

Candidates report three rounds over roughly 3-5 weeks. The loop starts with an Initial Screening, moves to Technical Discussions with hiring managers and team members, and ends with a Cultural Alignment stage.

Focus areas

• C#
• .NET (including.NET Core/.NET 6+)
• Machine Learning Pipelines (end-to-end)

Overview

Candidates describe the job as building software for physical machines, not web products. You would build desktop and Android control applications in C# on modern .NET, keep their data in SQL Server or PostgreSQL, and run ML models inside the machine software. To prepare well, you need to talk about all three layers in one conversation: the application architecture, the database under it and the model running inside it.

All of the reported technical questions ask about your own experience: coding in production, .NET Core and API design, preprocessing and deploying ML models, and testing for code quality. None of them is an algorithm puzzle. So the depth has to come from your own projects. For each system you mention, be ready to explain its layers, how dependency injection wires it together, how its API contract is shaped, which indexes the database relies on, and why you made each choice.

Candidates also report two groups of scenarios. On the systems side: designing a data schema for a new automation module, and handling migrations and performance tuning in a production database. On the ML side: a model you deployed and the validation problems you hit, keeping a model maintained in production, choosing between techniques such as regression and clustering for an automation task, and embedding a model in machine control software. Prepare one concrete, numbers-backed story for each scenario.

The final stage covers fit and alignment with the company's mission. Before it, learn what aseptic manufacturing involves, why contamination control and repeatability matter there, and what that means for software: careful releases, thorough testing and clear operator documentation. Candidates also report that unrestricted US work authorization and a drug test are part of the hiring process, so confirm both early.

The AST interview process

Candidates report the three stages below in this order, but only each stage's name and purpose are reported. The preparation under each stage is advice on getting ready, not a description of what happens in the room.

Initial Screening

Candidates report that the loop starts with an initial screening of basic qualifications and fit. Nothing more about its content is reported. Use it to show quickly that you meet the reported requirements: production C# on .NET Core/.NET 6+, SQL Server or PostgreSQL, ML pipelines you took from data to deployment, and Scrum/Agile delivery. Give one concrete example for each instead of a list of keywords. This is also the stage to confirm the eligibility conditions candidates report: unrestricted US work authorization and a drug test.

What to demonstrate

• Basic qualifications against the reported requirements: C#/.NET, a SQL database, end-to-end ML pipelines and Agile experience
• Overall fit for the role, which candidates give as the purpose of the screen

How to prepare

• Write a short background summary that names one C#/.NET system you built, one database you designed or tuned, and one ML model you deployed, each with a number (users, data volume, latency or machines).
• Write an honest one-line answer for any requirement you only partly meet, such as limited Android or PLC exposure, along with what you have done to close the gap.
• Confirm you meet the reported US work-authorization and drug-test conditions before the process goes further.
• Write two questions about the team's current R&D work and the machines your software would control.

Technical Discussions

Candidates report in-depth technical discussions with hiring managers and team members to assess technical skills. The reported technical questions are about your own experience, not puzzles, so prepare for discussion about real systems. For each project you bring, be ready to explain its architecture, data model, API, tests and deployment, and why you made each decision. Practice switching between the .NET application layer, the SQL Server or PostgreSQL schema under it and any ML model inside it, because the reported scenarios cover all three.

What to demonstrate

• Technical skills, assessed in discussion with hiring managers and team members
• Depth on the systems in your own background, including the reasons behind design choices
• How clearly you explain technical work, since this stage is reported as a discussion

How to prepare

• Pick one .NET 6+ service or desktop application you built and practice sketching its layers, DI registrations, data access and API contract from memory.
• Outline a schema for a new automation module: machines, recipes, batches and timestamped sensor readings. Include keys, the indexes your main queries need, and how you would run the migration on a live database.
• Prepare a complete walkthrough of one ML model you took to production: preprocessing, validation, how it is packaged into the application, and how it is monitored and updated.
• Prepare one production debugging or performance story with the before and after numbers.

Cultural Alignment

Candidates report that the last stage assesses cultural fit and alignment with the company's mission and values. No specific values are reported, so do not guess at a list. Prepare a grounded reason for wanting to build control software for life science manufacturing, and behavioral stories that show independent problem-solving, working with non-technical colleagues, and handling feedback and disagreement. The reported behavioral questions are about strengths and weaknesses, independent work under time pressure, gathering requirements with non-technical stakeholders, and motivation in a fast-moving R&D environment.

What to demonstrate

• Cultural fit, as candidates describe this stage
• Alignment with the company's mission, including whether your interest in automation for life science manufacturing is specific and credible

How to prepare

• Read up on what aseptic manufacturing involves and write three sentences on why software reliability matters in it.
• Prepare a STAR story for each reported behavioral question, each with a measurable result and something you changed afterwards.
• Prepare one story about a disagreement over requirements or design and how it was settled.
• Prepare one story about learning a new technology or domain quickly, since candidates describe the team as R&D.

Where candidates lose points

Answering the .NET Core and API design question with a list of buzzwords ("REST, DI, microservices") and no API you actually built.

Prepare one real service and walk through it concretely: routes, status codes, error bodies, DI lifetimes, async handling and versioning. Have one design decision you would change now, and why.

Ending the ML pipeline answer at training accuracy, with nothing on preprocessing parity, validation or how the model runs in production.

Cover the full path: leakage-safe preprocessing, a split by time, batch or machine where rows are correlated, preprocessing shipped and versioned with the model, a fallback for bad inputs, and drift monitoring with a defined process for promoting a retrained model.

Answering the testing question with a coverage percentage and nothing about how the code was made testable.

Explain the unit and integration layers you use, how hardware, clock and database dependencies sit behind interfaces, how integration tests run against a real SQL Server or PostgreSQL, and one bug your tests caught.

Struggling on the reported database scenarios: designing a schema for an automation module, or doing a migration and performance tuning on a live database.

Practice one schema out loud with keys and indexes for its main queries. Prepare a migration story that covers backward-compatible steps, backfills and rollback. Know how to read an execution plan and explain why a query chose a scan over a seek.

A vague motivation or culture answer that could apply to any software company.

Learn what aseptic manufacturing involves and connect it to how you work: careful releases, thorough testing and clear documentation for machine operators. Use a real story from your past as evidence.

Questions & practice

Collapse a redelivered event batch into per-aggregate high-water marks

Easy · hashing · at-least-once · aggregation

You drain a batch of up to 5,000,000 events, each (aggregate_id BIGINT, aggregate_version INT, event_type, payload). The log guarantees order within one aggregate only; the batch merges 64 partitions, and a relay failover has redelivered a range, so an older version for an aggregate can appear after a newer one. Given a map of last_applied_version per aggregate, produce the events worth applying, at most one per (aggregate_id, version), plus the count discarded. Target O(n) time. State the memory for 2,000,000 distinct aggregates and what you do when it does not fit.

Approach

1. One pass, one hash map from aggregate_id to the highest version kept, and a discard counter. An event whose version is at or below last_applied_version for its aggregate is dropped without further work, which is the whole reason the event carries its version rather than a delta. O(n) expected time, O(d) space in distinct aggregates.
2. Keep the maximum, never the last occurrence. The redelivered range means the final appearance of an aggregate in the batch can be an older version than one seen earlier in the same batch, so last-wins applies stale state over newer state and the projection regresses with no error anywhere.
3. Cost the memory instead of calling it large: an 8-byte key plus a 4-byte version is 12 bytes of payload, and an open-addressed table held at a 0.7 load factor costs roughly 17 bytes per entry before per-slot metadata, so 2,000,000 aggregates is tens of megabytes in a native layout and several times that in a runtime that boxes both key and value.
4. If the distinct set exceeds memory, partition on hash(aggregate_id) mod P and reduce each partition independently. Every event for one aggregate hashes to the same partition, so the per-partition result is exact and the merge is concatenation rather than a second reduction.

Follow-up

• The payload is a patch rather than a snapshot, so applying only the highest version loses the intermediate changes. What changes in your reduction?
• How do you detect that version 7 arrived while version 6 was never delivered, and what should the consumer do about the gap?

Archive a resource graph without breaking live references or recursing

Medium · graph traversal · topological order · tenant isolation

Resources reference other resources within a tenant; for the largest tenant the reference table holds up to 2,000,000 nodes and 8,000,000 edges. Archiving a resource must archive everything reachable from it that nothing outside the set still references, refuse when a live external referrer exists, and terminate when references form cycles, which they legitimately do. Produce the archive order and the refusal list, targeting O(V+E). Say what stops the traversal crossing a tenant boundary, and why recursion is the wrong control structure at this size.

Approach

1. Load the subgraph with the tenant predicate on both endpoints of the edge, not only on the side you started from. Scoping the left table alone is the classic cross-tenant leak: one mis-entered edge then pulls another tenant's resources into the traversal and, worse, into the archive.
2. Traverse iteratively with an explicit stack. A 2,000,000-node graph can hold a chain deep enough to exhaust a native stack in the low tens of thousands of frames, and that failure is a process crash rather than an error you can return.
3. Treat cycles as data rather than corruption: compute strongly connected components with Tarjan in O(V+E) using its own explicit stack, then condense. The condensation is a DAG, so a topological order over it gives the archive order, and every member of a component archives in one transaction because no order within a cycle is valid.
4. Decide refusals with reverse edges. A candidate is archivable only if every in-edge originates inside the candidate set, so build the transpose or count in-degrees restricted to the visited set, and emit each blocked resource with the id of the external referrer, which is the only part of the answer an operator can act on.

Follow-up

• The graph is read in one query and the archive writes a minute later. What can change in between, and how do you make the write safe?
• The candidate set is 400,000 resources. Is that one transaction, and if not, what does a half-finished archive look like to a reader?

Merge partitioned event streams into one ordered feed with bounded lateness

Hard · k-way merge · watermarks · out-of-order streams

The read-model service consumes 64 log partitions carrying about 4,000 events per second in total. Each partition is ordered within itself, but partitions drift by up to 30 seconds, and the activity feed must present a tenant's events in occurred_at order. Produce the merge. State its complexity, the buffer it requires in events and in bytes, what happens when one partition is idle, and what you do with an event that arrives after you have already emitted its position. Payloads average 1 KB.

Approach

1. Merge with a min-heap over the 64 partition heads keyed on (occurred_at, event_id): O(log P) per event and O(n log P) overall. The tie-break on event_id is what makes the output deterministic when two partitions carry the same millisecond, which matters because the feed is paginated and a non-deterministic order reorders pages under the reader.
2. Emitting the heap head is only correct once every partition has produced everything up to that timestamp, so the emit condition is a watermark: the minimum across partitions of the highest occurred_at seen, less the allowed lateness. Events are held until the watermark passes them, which is what turns individually ordered streams into a jointly ordered one.
3. Size the buffer from the lateness rather than guessing: 4,000 events per second times 30 seconds is 120,000 buffered events, and at 1 KB each about 120 MB of heap. That number is the real price of the ordering guarantee and belongs in front of whoever asked for it.
4. Handle the idle partition explicitly, because it fails the feed rather than corrupting it: a partition with no traffic never advances its own maximum, so the watermark freezes and output stops entirely. Either every partition emits a periodic idle marker carrying the broker's current time, or the watermark falls back to wall clock for a partition silent beyond a threshold.

Follow-up

• The lateness budget is raised to five minutes. What is the new buffer, and what besides memory changes?
• The consumer restarts. Where does it resume from, and what does the feed look like for the first 30 seconds?

Hold a per-tenant active cap against concurrent creates

Hard · write skew · isolation levels · constraints

A tenant on the standard plan may hold at most 50 resources with status='active'. The create handler runs SELECT count(*) FROM resource WHERE tenant_id = $1 AND status = 'active', compares to 50, then inserts. Two creates arrive 3 ms apart on different instances and the tenant lands at 51. Name the anomaly, say whether PostgreSQL 16 READ COMMITTED or REPEATABLE READ prevents it and why, then give an implementation that holds the cap at READ COMMITTED with the exact statements. Finally, say what changes when the cap is 'at most one running export per tenant' on job_run.

Approach

1. Name it: write skew. The two transactions read an overlapping set and write disjoint rows, so there is no row-level conflict for the engine to detect and each commit is individually legal.
2. Rule out the levels precisely. READ COMMITTED takes a fresh snapshot per statement and takes no lock on the counted rows, so both see 49. PostgreSQL's REPEATABLE READ is snapshot isolation: it removes non-repeatable reads and phantoms within the snapshot but still admits write skew, because the anomaly is not a re-read of a changed row, it is a read of a set that a concurrent transaction invalidates. Only SERIALIZABLE closes it, by tracking the read dependency and aborting one transaction with SQLSTATE 40001 — a guarantee that exists only if the application re-runs the whole transaction from the read.
3. Convert the set predicate into a single-row conflict: keep tenant.active_resource_count and run UPDATE tenant SET active_resource_count = active_resource_count + 1 WHERE tenant_id = $1 AND active_resource_count < 50 in the same transaction as the INSERT. Zero affected rows is the cap, returned as 409. The row lock serialises the decision at any isolation level, and contention is bounded to one tenant's row — which is also the fair-scheduling unit, unlike a global counter that would convoy every tenant behind one row.
4. State the cost you just took on: a counter is a second source of truth that can drift, so every path that changes status must adjust it inside the same transaction, and a periodic reconciliation has to exist, with resource_revision as the authority for what the count should have been.

Follow-up

• A resource moves from archived back to active. Which statements change, and what breaks if the counter update and the status change land in different transactions?
• The cap becomes plan-dependent and a plan can change mid-month. Where does the number 50 live, and who reads it?

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

1. 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.
2. 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.
3. 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.
4. 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?

How have you handled data preprocessing and model deployment in machine learning pipelines?

Medium · Technical Competency and Domain Experien

Approach

1. Walk through the pipeline in order: ingestion (where the data comes from and its quality problems), cleaning (missing values, outliers, aligning units and timestamps), feature engineering, training, validation, packaging, deployment and monitoring.
2. Name the leakage traps. Fit scalers, imputers and encoders on the training split only. When rows from the same batch, run or machine are correlated, split by time, batch or machine instead of randomly, or the validation score will be inflated.
3. Explain how preprocessing ships with the model so serving matches training. One option is a single pipeline object; another is exporting to a portable format such as ONNX and running it from C#. Either way, version the model and its preprocessing together.
4. Describe deployment and validation: offline or shadow comparison against the current model, acceptance thresholds agreed before the run, a fallback when inference fails or inputs are out of range, and latency limits when the model runs inside machine control software.
5. Finish with upkeep, which is one of the reported scenarios: monitor input drift and the distribution of predictions, log inputs and outputs for retraining, and promote a new version only after it passes the same validation as the first.

Follow-up

• Describe a model you deployed and the hardest problem you hit while validating it.
• How do you update a model that is running in production without disrupting the software around it?
• How would you detect that live inputs have drifted from the training data?

Explain your approach to ensuring code quality and maintainability through unit and integration testing.

Medium · Technical Competency and Domain Experien

Approach

1. Lay out the layers: fast unit tests for domain logic (xUnit or NUnit, with fakes from Moq or NSubstitute), integration tests against a real database and the real HTTP pipeline (WebApplicationFactory with a containerized SQL Server or PostgreSQL), and a small set of end-to-end checks.
2. Explain how you design for testability. Depend on interfaces injected through DI, put the clock, file system and any hardware or PLC communication behind abstractions, and keep side effects at the edges of a layered architecture.
3. Make tests deterministic: no sleeps, a controllable clock, separate data for each test, and a clear policy for flaky tests (quarantine and fix them, never rerun until green).
4. Connect it to maintainability beyond tests: code review, analyzers and nullable reference types, CI that blocks merges on failing tests, and coverage treated as a way to find untested code, not as a target.
5. Give one concrete example: a bug a test caught before release, or a regression that slipped through and the test you added afterwards.

Follow-up

• How would you test code that talks to a PLC or other hardware you cannot run in CI?
• How do you test a database migration before it reaches production?
• What would you not bother to unit test, and why?

p99 jumped on one listing filter while p50 stayed flat

Medium · query planning · composite index · percentiles

After a release that added an owner_user_id filter to the resource listing, p99 rose from 90 ms to 1.9 s while p50 stayed at 40 ms. Traffic and row counts are unchanged. resource carries the index (tenant_id, status, updated_at DESC, resource_id DESC). The new query filters tenant_id and owner_user_id, orders by updated_at DESC, resource_id DESC, and takes 20 rows. On PostgreSQL, explain the shape of the regression, prove it from a query plan, and give the index you would add.

Approach

1. Start from the shape. A flat p50 with a moved p99 means a subset of requests changed cost, not all of them, so the first job is naming the subset. Bucket the endpoint's latency by the tenant's row count; the natural hypothesis is that large tenants are a small share of requests and all of the tail.
2. Get the plan for the new query on a large tenant with EXPLAIN (ANALYZE, BUFFERS). Expect an index scan over the tenant's range, a filter discarding most of it, then a Sort feeding the Limit, possibly reporting Sort Method: external merge Disk. Read actual rows on the scan node, not estimated.
3. Explain why the existing index cannot serve it. A composite B-tree is seekable only as a left prefix, and with no equality predicate on status the scan cannot treat updated_at as an ordering, because rows in the tenant's range are ordered by status first. Everything matching must be read and sorted before LIMIT 20 can apply, so a tenant with 400,000 rows pays 400,000 rows to return 20.
4. Add (tenant_id, owner_user_id, updated_at DESC, resource_id DESC). Equality on the first two columns leaves the index ordered by updated_at within that pair, so the plan becomes an index scan that stops after 20 rows with no Sort node. PostgreSQL can scan a B-tree backwards, so the DESC markers matter only if the two sort columns ever disagree in direction; keeping them explicit documents the order the keyset cursor depends on.

Follow-up

• The endpoint paginates with OFFSET. What does page 500 cost with your index, and what does the keyset version cost?
• How would you have caught this before release, given that a 10,000-row seed database produces the same plan shape at an unnoticeable cost?

Suggested practice checklist

The plan follows the three reported stages and the requirements candidates describe: C# and .NET 6+, SQL Server or PostgreSQL, end-to-end ML pipelines and testing. Each reported question is practiced on the day its subject is covered, and the last day is a full mock.

Day 1 — Map the loop and inventory your stories

• Write a short background summary covering the reported requirements: production C#/.NET, a SQL database, an ML model in production and Agile delivery.
• List every production system you could talk about. Pick the one you know best and write its stack, scale, deployment process and one incident in STAR form.
• Practice your strengths-and-weaknesses answer out loud, with evidence for each strength and a mitigation plan for the weakness.

Deliverable: A background summary, a one-page sheet on your main production system with an incident story, and a recorded strengths-and-weaknesses answer.

Day 2 — C# and .NET Core API depth

• Sketch the API you know best: routes, middleware order, DI registrations with lifetimes, and error responses as ProblemDetails.
• Write a short note on the captive-dependency bug, blocking on .Result, and how you would version a live API.
• Practice the .NET Core and API design answer out loud, then answer its follow-up on scoped services inside singletons.

Deliverable: An architecture sketch of one API, a one-page note on DI and async pitfalls, and a recorded answer to the API design question.

Day 3 — Database design, migrations and tuning

• Outline a schema for a new automation module (machines, recipes, batches, timestamped readings) with keys and indexes for its main queries.
• Write the steps for a backward-compatible migration on a live table: add column, backfill, switch reads, remove the old column.
• Work the practice questions on enforcing a per-tenant cap under concurrent inserts and on joins that fan out a page, and explain the locking and indexing behind your answers.

Deliverable: A schema with an index list, a written migration plan, and two solved SQL practice questions.

Day 4 — End-to-end ML pipeline

• Write out one model you took to production, stage by stage: ingestion, cleaning, features, split strategy, validation, packaging, deployment and monitoring.
• List the leakage risks in that pipeline and how you avoided each one, and note when you would choose regression or clustering for an automation task.
• Practice the preprocessing and deployment question out loud, including how you would update the model in production without disrupting the software it runs in.

Deliverable: A one-page pipeline walkthrough with a leakage checklist, and a recorded answer to the ML pipeline question.

Day 5 — Testing, quality and production debugging

• Write your test strategy for a control application: which unit tests use fakes, which integration tests hit a real database, and how hardware access sits behind an interface.
• Practice the testing question out loud with one example of a bug a test caught.
• Work the practice question on a p99 latency jump while p50 stayed flat, and write the order in which you would investigate.

Deliverable: A one-page test strategy, a recorded testing answer, and a written debugging sequence for the latency problem.

Day 6 — Behavioral stories for Cultural Alignment

• Write STAR stories for independent work under time pressure and for gathering requirements with non-technical stakeholders, each with a measurable result.
• Write your motivation answer, tied to control software for life science manufacturing and supported by one story from your past.
• Practice all three out loud, then answer the follow-up on a requirement that changed late.

Deliverable: Three written STAR stories and a recorded run of the three behavioral answers.

Day 7 — Full mock of the three stages

• Have a friend run a mock screen using your background summary and eligibility answers.
• Run a technical discussion mock: production coding, .NET API design, the ML pipeline and testing, with follow-ups from this guide.
• Run a behavioral mock with all four reported behavioral questions, then write two questions to ask about the team's R&D work.

Deliverable: Notes from three mock sessions listing the answers that need work, and two questions to ask the team.

Behavioural preparation

Candidates report that the final stage covers cultural fit and alignment with the company's mission. The reported behavioral questions focus on self-awareness, solving problems independently under time pressure, working with non-technical stakeholders on requirements, and motivation in an R&D setting. Back every story with a measurable result and something you changed afterwards.

1. Tell me about a production issue you debugged and what you changed so it would not recur.
2. Tell me about a disagreement over a design or requirement and how it was resolved.

Can you describe your experience with coding in production environments?

Medium · Technical Competency and Domain Experien

Approach

1. Lead with one system you owned end to end: what it does, who uses it, its stack (for example an ASP.NET Core service on .NET 6+ backed by SQL Server or PostgreSQL), and how often you ship to it. One system described in detail beats a list of five.
2. Walk through the production lifecycle: how code gets from your branch to production (CI build, tests, review, release), how you roll back, and which telemetry tells you it is healthy (structured logs, metrics, alerts).
3. Bring one incident in STAR form. Cover the symptom, how you narrowed it down (logs, a reproduction, a profiler, a query plan), the fix, and the safeguard you added so it cannot happen again, such as a regression test or an alert.
4. If you have shipped software that runs on hardware or at customer sites, name the constraint that created. For example, you cannot patch a machine in the middle of a run, so a release needs validation and a rollback plan before it goes out.
5. Be honest about scale and ownership: users, request rates, data volume or number of machines, and which parts were yours and which were the team's.

Follow-up

• What was the worst production failure you were responsible for, and what changed afterwards?
• How do you release a change to software that is already running on a customer's machine?
• How did you confirm the fix worked, rather than the symptom just going quiet?

What is your experience with.NET Core and modern API design?

Medium · Technical Competency and Domain Experien

Approach

1. Base the answer on one API you built on .NET Core/.NET 6+. Cover controllers or minimal APIs, the middleware pipeline (authentication, exception handling, logging), and how requests reach your services through the built-in DI container.
2. Show you understand DI lifetimes: singleton, scoped (one per request) and transient. Explain the captive-dependency bug, where a singleton holds on to a scoped DbContext, and mention binding configuration with IOptions<T>.
3. Cover contract design: resource-oriented routes, correct status codes, input validation with consistent ProblemDetails error bodies, pagination, versioning by URL or header, and OpenAPI documentation that clients can generate code from.
4. Handle async correctly: async all the way down, a CancellationToken passed through to EF Core or Dapper calls, and never blocking on .Result or .Wait(), which can exhaust the thread pool under load.
5. Add reliability and security: idempotency keys so a retried POST is not applied twice, timeouts and retries with backoff on outbound calls through IHttpClientFactory, and authentication plus authorization policies on every endpoint.

Follow-up

• What happens if you inject a scoped service into a singleton, and how would you catch it?
• How would you version an API that deployed machines or clients already depend on?
• Have you moved code from .NET Framework to .NET 6+, and what broke?

Can you walk me through your personal strengths and weaknesses?

Medium · Behavioral and Soft Skills

Approach

1. Choose two or three strengths that match the reported requirements, such as production C#/.NET depth, taking an ML model all the way to deployment, or turning vague requests into a clear spec. Back each one with a specific result.
2. Name a real weakness that does not rule you out for the role. Show its cost with one example, then spend most of the answer on how you manage it and the evidence that it is improving.
3. Avoid disguised strengths such as "I work too hard". Also avoid naming a gap in a must-have skill, such as SQL tuning, without a concrete plan you have already started.
4. Keep the structure tight: strength and evidence; weakness, example, what you do about it, and where it stands now.

Follow-up

• Can you give an example where that weakness caused a problem?
• What would your last manager say you should work on?
• How would that strength show up in your first project on this team?

Describe a time you had to work independently to solve a complex, time-sensitive problem.

Medium · Behavioral and Soft Skills

Approach

1. Use STAR. In the Situation, establish both constraints the question names: why the problem was complex (several interacting causes, an unfamiliar system) and why time mattered (stopped equipment, a release date, a waiting customer).
2. Spend most of the answer on your actions: how you triaged, which hypotheses you ranked and tested, which tools you used (logs, debugger, profiler, query plans), and when you chose a safe workaround over a full fix.
3. Show independence without going dark: you worked alone, but you kept stakeholders updated on status and escalated decisions that were above your authority.
4. End with a measurable result and what you put in place afterwards, such as a runbook, a test or monitoring, so it does not happen again.
5. Avoid a hero story where you skipped review or testing to ship faster. In software that controls machines, that kind of shortcut is a liability, not a strength.

Follow-up

• What would you have done if your workaround had failed?
• At what point did you decide to ask for help, or why didn't you?
• What did you leave behind so the next person could handle the same problem?

How do you handle collaboration when defining project requirements with non-technical stakeholders?

Medium · Behavioral and Soft Skills

Approach

1. Describe a concrete method. Talk to the actual users (for machine software, for example operators or process engineers), watch the current workflow, and turn their needs into examples and acceptance criteria they can confirm.
2. Use artifacts that non-engineers can react to: screen mockups or a clickable prototype of the UI, a short written spec stating what is in and out of scope, and a demo at the end of each sprint in a Scrum process.
3. Show how you handle conflict and change. Separate the underlying need from the solution being asked for, state trade-offs in cost and schedule, and send scope changes through the backlog instead of absorbing them quietly.
4. Tell one STAR story where this process caught a misunderstanding early, or where it failed and what you changed afterwards.

Follow-up

• Tell me about a requirement that changed late. How did you handle it?
• How do you explain a technical constraint to someone who just wants the feature?
• How do you confirm you built the right thing after delivery?

What motivates you to contribute to a shared vision in a fast-paced R&D environment?

Medium · Behavioral and Soft Skills

Approach

1. Be specific about why this work interests you: software that controls physical machines, the life science manufacturing context, and R&D work that builds new capability. Generic enthusiasm could apply to any employer.
2. Back up the motivation with your history: a project where you chose harder, domain-heavy work, learned a new field quickly, or picked up a new technique and shipped it.
3. Address the fast-paced part honestly. Explain how you handle shifting priorities and incomplete requirements while still testing and documenting your work.
4. Show you can commit to a shared direction that was not your idea. Give an example of backing a team decision you had argued against, and delivering on it.

Follow-up

• What do you know about what AST's machines are used for?
• Tell me about a time priorities changed partway through a project.
• What kind of work would make you want to leave a role like this?