Applied Energy Systems Software Engineer Interview Guide

The 60-second brief

Applied Energy Systems builds control systems and software for gas delivery equipment used in industrial settings, including high-purity gas delivery. A Software Engineer there works on code that runs on the company's own specialized hardware. Candidates describe the job as covering the whole development cycle: writing, testing and debugging control software, taking part in code reviews, contributing to architecture documentation and, now and then, supporting systems already installed in the field. Much of the day is spent alongside mechanical and electrical engineers, turning what the physical equipment can and cannot do into software behavior.

This guide is for candidates applying to the Software Engineer role at Applied Energy Systems, which candidates describe as entry to mid level and based on site in Malvern, PA. It covers the two reported rounds, the thirteen reported questions (real-time processing on limited resources, software and hardware debugging, language choice for control systems, documentation, industrial protocols, legacy code, performance versus safety, technical debt, and the behavioral set) and a seven-day plan.

Candidates report two rounds, an Initial Screening followed by Technical Deep Dives, over roughly two to four weeks. The reported questions mix hardware-adjacent engineering topics with collaboration and decision-making stories.

Focus areas

• Software Engineering
• Programming Fundamentals
• Problem Solving

Overview

The reported questions lean toward software that touches physical equipment. Candidates mention handling real-time data on limited resources, tracking down faults that sit between software and hardware, choosing a language for control systems, and experience with Modbus or similar industrial standards. That shifts preparation away from puzzle-style computer science and toward being able to describe, concretely, how a program behaves when memory is small, a sensor goes quiet or a bus times out.

Candidates describe the technical evaluation areas as embedded programming and hardware interfaces, a basic grasp of control theory (feedback loops and stability), real-time operating systems, memory management and hardware abstraction layers, plus architecture topics such as modular, independently testable components, error handling for environments where downtime is costly, distributed design and secure communication. The listed requirements are C/C++, Git and software testing methods, with PLC programming and industrial automation tools as extras. The topic tags on the page also include data structures and algorithms, so keep ordinary coding fluency warm.

The behavioral questions candidates report are about technical disagreement, explaining complex ideas to non-technical people, pivoting when requirements change, designing a system from scratch, and why the industrial gas and energy sector. Candidates also describe a tip to explain why you chose a technology, not only what you built. Prepare stories where the trade-off, the failure state you guarded against and your own decision are all visible.

Candidates mention a technical assessment on writing readable, efficient code, sometimes in hardware-interfacing scenarios, without saying when it happens. Practise writing small C/C++ routines with tests (a buffer, a frame parser, a simple controller) so that scenario feels familiar, and keep your hardware-facing projects ready to discuss in detail.

The Applied Energy Systems interview process

Candidates report two stages. The descriptions below say what each stage covers as reported and how to prepare for it, without tying specific questions to a specific round.

Initial Screening

The Initial Screening is reported as the stage that looks at your background and fit for the role. Treat it as the point where your resume gets explained: candidates are advised to highlight projects that involved hardware interaction and to say why they chose particular technologies. Walk through your experience with the hardware boundary in mind: what device or bus your code talked to, what language it was in, and what you owned. Keep the sector-motivation answer ready too, and bring a question or two on how the software team works with the mechanical engineers.

What to demonstrate

• Background and experience relevant to a Software Engineer role
• Fit for the role, as the stage is described by candidates
• How clearly you explain why you chose the technologies in your past projects

How to prepare

• Choose two projects that touched hardware or a physical process. For each, write the interface (bus, protocol or driver), your own contribution and one fault you tracked down.
• For every technology on your resume, write one sentence on why it was chosen over the alternative, and what it cost.
• Prepare an honest answer to why industrial gas and energy work interests you, tied to something you have actually built or studied.
• Write two questions about how the software team and the mechanical team hand work to each other.

Technical Deep Dives

Technical Deep Dives are described as detailed technical discussions about engineering challenges in the energy sector. Reported topics on this loop include embedded programming and hardware interfaces, feedback loops and stability, RTOS and memory management, modular and testable design, error handling, and industrial protocols such as Modbus. Approach it by stating the constraint first (memory, deadline, safety state), then the design, then what happens on failure. Candidates are told to mention edge cases and failure states in every answer, so build that habit.

What to demonstrate

• Technical depth in C/C++ and the hardware-facing topics candidates list
• Reasoning about trade-offs and why one approach beat another
• Design for failure: edge cases and failure states in the code you describe
• Structured problem solving, with STAR suggested for describing multi-layered problems

How to prepare

• Write a fixed-size ring buffer in C++ with no dynamic allocation after start-up, and test the full, empty and wrap-around cases.
• Summarize Modbus in your own words: master and slave roles, registers and coils, common function codes, RTU versus TCP framing, checksums, timeouts and exception responses.
• Review RTOS basics you can explain without notes: task priorities, mutexes, priority inversion, interrupts handing work to tasks, and why heap allocation is avoided in loops.
• Prepare a step-by-step checklist for a fault between software and hardware, and one real story where you used it.

Where candidates lose points

Answering real-time and hardware questions with web-service habits: dynamic allocation in the hot path, retries without bounds, no deadline stated

Begin every such answer by stating the memory budget, the deadline and what happens when the deadline is missed. Then describe a bounded design, such as a fixed buffer, a bounded loop and a defined overload policy.

Describing a software and hardware debugging story as 'I kept trying things until it worked'

Narrate a sequence: confirm what changed, check power, wiring and configuration, capture traffic or signals with a logic analyzer or logs, and isolate software from hardware with a known-good test. Name the evidence that separated cause from trigger.

Saying 'safety is important' with no concrete failure state behind it

For each design you describe, name one failure (sensor timeout, lost frame, buffer full, watchdog reset) and the safe default the code falls back to. Say how you tested that path.

Claiming Modbus or PLC experience that dissolves under one detail question

State precisely what you have used (RTU or TCP, master or slave, which register types) and what you have only read about. Prepare the basics of framing, function codes, checksums and timeouts so the honest answer is still a good one.

Behavioral stories told as 'we', with no personal decision and no result

For each story write down the one decision that was yours, the alternative you gave up and a number or observable outcome. Rehearse the disagreement, pivot and design-from-scratch stories aloud until each has all three.

Questions & practice

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?

Canonicalise a request body into a stable idempotency fingerprint

Medium · parsing · canonicalisation · hashing

idempotency_key.request_fingerprint is a SHA-256 over the method, path and canonicalised body, and a retry whose fingerprint differs must be rejected with 422 rather than served the stored response. Write the canonicaliser. Bodies are JSON up to 256 KB nested at most 32 levels; clients vary key order, whitespace and unicode escaping, and some send 64-bit ids as JSON numbers. Produce a deterministic byte string such that semantically identical bodies match and any semantic difference does not. State your complexity and name two normalisations you refuse to perform.

Approach

1. Parse once into a tree, then re-serialise under fixed rules: object keys sorted, array order preserved, one escaping convention, no insignificant whitespace. Parsing is O(n) and sorting keys is O(k log k) per object, so O(n log n) overall with O(depth) stack, and the 32-level cap is enforced during parsing because hostile nesting is how a canonicaliser becomes a stack overflow.
2. Sort keys by their UTF-8 bytes and say why the obvious implementation is wrong in some runtimes: a default string comparison that orders by UTF-16 code units places surrogate pairs, meaning code points from U+10000 up, below U+E000 to U+FFFF, which is not UTF-8 byte order, so two services written in different languages disagree on the same document.
3. Do not re-encode numbers through a double. IEEE-754 binary64 represents integers exactly only up to 2^53, so normalising a 19-digit id through a float changes it, and 1 against 1.0 cannot be reconciled without deciding whether they are the same value. Preserve the literal token, and require ids as strings at the API boundary if you want them comparable.
4. Reject duplicate keys rather than picking one. JSON permits them and parsers disagree, most keeping the last, so any choice you make ties the fingerprint to a parser detail that the code handling the request does not necessarily share.

Follow-up

• A client sends the same logical request with an extra field your API ignores. Same key, different fingerprint, so you return 422. Is that the right answer?
• Where does the fingerprint get computed relative to request decompression and the body-size limit?

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?

Find version gaps and relay lag with window functions

Medium · window functions · event tables · frame semantics

outbox_event holds event_id, aggregate_type, aggregate_id, aggregate_version, event_type, payload, status ('pending','published','dead'), attempts, created_at, published_at. A projection is missing rows and you must decide whether the relay skipped events or the consumer dropped them. Write three queries over the last seven days: one listing every aggregate_id whose published aggregate_version sequence has a hole, one giving per-day counts with a running total, and one returning the newest published event per aggregate. For each, say where the window function is evaluated relative to WHERE and LIMIT. PostgreSQL 16.

Approach

1. Gaps: compute lead(aggregate_version) OVER (PARTITION BY aggregate_id ORDER BY aggregate_version) in a subquery, then filter next_version <> aggregate_version + 1 in the outer query. Window functions are evaluated after WHERE, GROUP BY and HAVING and before the outer ORDER BY and LIMIT, so the predicate cannot sit in the same WHERE clause and PostgreSQL 16 has no QUALIFY.
2. Say what the seven-day filter does to the answer: it truncates every partition, so the first row per aggregate has no predecessor inside the window and a hole spanning the boundary is invisible. Widen the window, or join to resource.version as the authority for the true maximum.
3. Running total: SELECT date_trunc('day', created_at) AS d, count() AS n, sum(count()) OVER (ORDER BY date_trunc('day', created_at) ROWS UNBOUNDED PRECEDING). An aggregate inside a window call is legal because grouping runs before windowing. The grouping key is unique per row here so ROWS and RANGE agree, but write the frame anyway — over ungrouped rows with tied timestamps the default RANGE frame pulls in every peer row and the total jumps.
4. Newest per aggregate: DISTINCT ON (aggregate_id) ... ORDER BY aggregate_id, aggregate_version DESC is the cheap PostgreSQL-only form when an index matches that order; row_number() OVER (PARTITION BY aggregate_id ORDER BY aggregate_version DESC) = 1 is the portable form and needs a subquery for the same evaluation-order reason as the gap query.

Follow-up

• Relay failover redelivers events. Does a duplicate break the gap query, and how would you detect one from this table alone?
• Turn the gap check into a continuous monitor rather than a query someone runs after an incident. What does it watch?

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?

Which programming languages do you prefer for control systems, and why?

Medium · Technical Proficiency and Domain Knowled

Approach

1. Frame the answer by constraints first: hard or soft real-time, available RAM and flash, whether the code runs on a microcontroller, an embedded Linux box or a PLC, any certification needs, and the language the existing codebase and team already use.
2. C for microcontroller firmware: predictable timing, no hidden allocation, direct register access and a compiler for nearly every target. Name its risks (undefined behaviour, buffer overflows) and the mitigations: a coding standard such as MISRA C, static analysis, and strict compiler warnings.
3. C++ with a restricted subset for larger controllers: RAII, strong types for units and registers, templates and constexpr for zero-cost hardware abstraction layers, usually with exceptions and RTTI disabled and no heap allocation after start-up. The requirements candidates describe for this role list C/C++, so be ready to defend either.
4. On the PLC side, the IEC 61131-3 languages (ladder diagram, structured text, function block diagram); PLC programming is listed as a nice-to-have for this role. Python fits test harnesses, data tooling and HMIs, not a hard real-time loop, because interpreter overhead and garbage collection add unpredictable pauses.
5. Close with a concrete project: the language I used, the constraint that decided it and one thing I would choose differently now. Rust can come up as a memory-safe option with no garbage collector; weigh that against toolchain support for the target and team familiarity.

Follow-up

• Would you allow dynamic memory allocation inside a control loop? Why or why not?
• Why are exceptions and RTTI often disabled in embedded C++, and how do you handle errors instead?
• How would you introduce a new language into an existing C codebase without destabilising it?

Explain how you document your code for long-term maintainability in an industrial setting.

Medium · Technical Proficiency and Domain Knowled

Approach

1. Comments explain why, not what: units, valid ranges, the safety reason behind a limit, and references to the datasheet section or register map a value comes from. A magic number such as a timeout or a pressure threshold gets a named constant and a comment giving its source.
2. API-level docs (for example Doxygen headers) for every public function: preconditions, units, return and error codes, and whether it is safe to call from an ISR or from multiple threads.
3. Design documents that outlive the code: state machine diagrams for the controller, interface documents with register maps and message tables shared with the electrical and mechanical engineers, and short decision records saying why a choice was made and what was rejected.
4. Traceability through version control: commit messages that reference the requirement or ticket, tagged releases whose version matches what is flashed on the unit, and a changelog that field support can read.
5. Keep docs next to the code so they change in the same review, generate API docs in CI, and treat tests as executable documentation of expected behaviour, including the failure states.

Follow-up

• How do you stop documentation drifting from the code over years of changes?
• What belongs in a code comment, a commit message and a design document respectively?
• How would you document a safety interlock so the next engineer does not remove it by accident?

What experience do you have with communication protocols like Modbus or similar industrial standards?

Medium · Technical Proficiency and Domain Knowled

Approach

1. Answer honestly first: which protocols I have used, on what hardware and in which role (master or slave, writing a driver or configuring a tool). If Modbus is new to me, name the closest experience (another serial or fieldbus protocol) and show I know how Modbus works.
2. Modbus basics: request-response between one client (master) and servers (slaves). Modbus RTU runs over serial lines, commonly RS-485, with frames of slave address, function code, data and a CRC-16, delimited by a silent interval of 3.5 character times. Modbus TCP uses port 502 and an MBAP header (transaction id, protocol id, length, unit id) instead of the CRC.
3. Know the data model and function codes: coils and discrete inputs are bits, holding and input registers are 16-bit. 01 and 02 read coils and discrete inputs, 03 and 04 read holding and input registers, 05 and 06 write a single coil or register, 15 and 16 write multiple. An exception response sets the high bit of the function code (0x80 added) and carries a code such as 02 illegal data address.
4. Name the classic pitfalls: off-by-one between the 1-based 40001-style documentation numbering and the 0-based address on the wire; 32-bit values split across two registers with vendor-specific word order; timeouts and retries on a shared bus; and the absence of any built-in authentication or encryption.
5. Mention adjacent industrial standards only if I can discuss them, for example OPC UA, EtherNet/IP or CAN, and say when I would pick one over Modbus (richer data model and built-in security in OPC UA versus Modbus's simplicity and broad device support).

Follow-up

• How do you read a 32-bit float from two holding registers when the device's word order is not documented?
• One slave on an RS-485 bus times out intermittently. What do you check, in what order?
• Modbus TCP has no authentication. How would you protect it on a plant network?

How do you approach optimizing a legacy codebase that is critical to production?

Medium · Problem-Solving and System Design

Approach

1. Measure before touching anything: define the metric that matters (control-loop period, CPU load, memory, response latency), record a baseline, and profile to find the actual hotspot, using a profiler on a host build or timers and GPIO toggles on the target.
2. Pin current behaviour down before changing it: characterization (golden-master) tests that record outputs for captured real inputs, plus unit tests around the seam I am about to change. Without them, every optimisation risks silently changing control behaviour.
3. Go after the biggest wins first: algorithmic changes (removing an O(n^2) search, avoiding repeated parsing or copying), then data layout and allocation, then compiler flags, and micro-optimisation last. One change per commit so each gain and each regression is attributable.
4. Ship like the code is production-critical: bench or hardware-in-the-loop testing, a staged rollout to a few units, the previous firmware kept ready for rollback, and monitoring of the same metric in the field to confirm the gain.
5. Prefer incremental refactoring over a rewrite: a rewrite throws away years of fixed edge cases. If a module really must be replaced, swap it piece by piece behind a stable interface and run old and new side by side where the hardware allows.

Follow-up

• How do you optimise a module that has no tests and whose original author has left?
• Your optimisation changes floating-point results in the last few digits. Is that acceptable, and how do you decide?
• At what point would you argue that a rewrite is justified?

Describe a situation where you had to balance performance with safety requirements.

Medium · Problem-Solving and System Design

Approach

1. Set the scene with the hazard and the numbers: what could go wrong physically, what the safe state is (for example a valve closed or an output de-energised), and the performance target that was in tension with it, such as a loop deadline or throughput.
2. Explain the options considered and why safety checks were never the thing removed. The usual move is to make them cheaper, not to skip them: precompute limits, check interlocks every cycle in the fast path, and push non-safety work (logging, UI updates, communications) to lower-priority tasks.
3. Show defence in depth: input validation on sensor data (range, rate-of-change, stale-data timeouts), a watchdog that forces the safe state if the loop stalls, fail-safe defaults on start-up and on communication loss, and hardware interlocks that do not depend on the software.
4. Quantify the result with worst-case figures, not averages: worst-case execution time or the maximum observed loop time before and after, and how the safety path was verified, for example by fault injection on the bench.
5. Close with ownership: who agreed the trade-off, how it was documented, and what I would do differently. Structure it as situation, task, action, result so the reasoning is not buried in background.

Follow-up

• What would you do if the safety check alone pushed the loop past its deadline?
• How did you prove the safety path still worked after the performance change?
• Who had the authority to accept the trade-off, and how was it recorded?

How do you manage technical debt while trying to meet tight project deadlines?

Medium · Problem-Solving and System Design

Approach

1. Make debt visible and concrete: every deliberate shortcut gets a ticket that states the cost of leaving it (slower changes, a known failure mode, manual steps for field support), so it can be prioritised against features rather than argued about vaguely.
2. Classify before deferring: debt on a safety path or in input validation does not get deferred; cosmetic or internal-structure debt can. Say explicitly which kind you took on to meet the deadline.
3. Protect the shortcut so it is safe to revisit later: tests around it, a named constant or config switch instead of scattered hard-coded values, and a comment that links to the ticket.
4. Pay it down in the normal flow: refactor code you are already changing, reserve a fixed slice of each iteration for debt, and cut feature scope before cutting tests or reviews when the deadline is tight.
5. Give one real example: the deadline, the shortcut taken, how it was tracked, and when and how it was paid back, or what it cost when it was not.

Follow-up

• How do you convince a manager to fund a refactor that ships no visible feature?
• Tell me about debt you decided never to pay back. Why was that the right call?
• How do you stop temporary shortcuts from becoming permanent?

Describe your process for debugging a complex integration issue between software and hardware components.

Medium · Technical Proficiency and Domain Knowled

Approach

1. Start by pinning down the symptom: how often it happens, on which units, with which firmware and hardware revisions, and what changed recently (a firmware build, wiring, a config value, a replaced sensor). Write the timeline down before forming a theory.
2. Split the chain into layers and test each boundary on its own: physical (power, grounding, cabling, RS-485 termination), link (baud rate, parity, stop bits, timing), protocol (framing, CRC, register addresses, byte order) and application logic. A logic analyzer or oscilloscope on the line shows whether the bytes on the wire match what my code thinks it sent.
3. Isolate my code from the device by substituting known-good parts one at a time: a loopback test, a vendor or PC-based master tool talking to the device, or my software talking to a simulator. Whichever side still fails is where the fault lives.
4. For intermittent faults, suspect timing: ISR priority, buffer overruns, missed deadlines, watchdog resets. Printing inside an ISR changes the timing, so I log timestamped events into a RAM ring buffer or toggle a GPIO and measure it on the scope.
5. Separate trigger from root cause, then lock the fix in with a regression test (hardware-in-the-loop or a simulated device that replays the captured traffic) and write up the failure signature so field support can recognise it.

Follow-up

• The fault disappears as soon as you add logging. How do you instrument it without changing the timing?
• How would you tell a noisy cable or missing termination apart from a firmware timing bug?
• How do you confirm the fix holds on units already deployed in the field?

Suggested practice checklist

The plan follows the reported topics: hardware-facing C/C++ and real-time work early in the week, then debugging, protocols and safety, maintainability and legacy code, and finally the behavioral stories. Each day ends in something you can reuse in the interview.

Day 1 — Map the loop and inventory your hardware experience

• Write out the two reported rounds, Initial Screening and Technical Deep Dives, and list what you would be asked to show in each.
• List three projects that touched hardware or a physical process, and for each write the interface, your role, the technology choices and why, and one fault you diagnosed.

Deliverable: A one-page loop map plus a three-project inventory with the technology choice rationale written next to each.

Day 2 — Real-time work on a constrained target in C++

• Write a fixed-size ring buffer in C++ with no allocation after construction, and unit tests for empty, full, wrap-around and overflow-drop behavior.
• Write a spoken answer for handling real-time data in a resource-constrained environment: budget, data path, overload policy, how you measured.

Deliverable: A tested ring buffer source file and a written answer outline that cites its overflow policy.

Day 3 — Industrial protocols and the software and hardware debugging method

• Write a one-page Modbus summary: roles, registers and coils, common function codes, RTU versus TCP framing, checksum, timeouts and exception responses.
• Write a layered debugging checklist for a fault between software and hardware, then attach a story of yours to it.
• Write an honest account of your protocol experience, separating what you used from what you only studied.

Deliverable: A Modbus cheat sheet, a debugging checklist with one real story mapped onto it, and a short protocol-experience statement.

Day 4 — Control loops, safety and language choice

• Write a small C++ simulation of a proportional controller driving a simple process, and add a test showing the fail-safe state when the sensor input stops.
• Write your answer on which language you would choose for a control system and why, covering C versus C++, determinism and memory use.
• Write a STAR outline for a time you balanced performance with safety.

Deliverable: A controller simulation with a sensor-timeout test, a language-choice answer with trade-offs, and one performance-versus-safety story outline.

Day 5 — Maintainability, legacy code and technical debt

• Take one old module of yours, add characterization tests that pin down current behavior, then refactor one function behind them.
• Write a documentation header and a short design note for that module, aimed at an engineer who will maintain it for years.
• Write a triage list for technical debt under a deadline: what you fix now, what you log, and how you justify the choice.

Deliverable: A tested and documented module, a design note and a ranked debt list with reasons.

Day 6 — Behavioral stories: design, disagreement and pivot

• Write STAR outlines for designing a system from scratch, a technical disagreement and a requirements pivot, each with your own decision and a measurable result.
• Rehearse each story aloud, then cut anything that is not the situation, your decision or the result.

Deliverable: Three STAR outlines with decisions and results marked, each rehearsed aloud.

Day 7 — Stakeholders, motivation and a full rehearsal

• Write an explanation of a feedback loop or buffer overflow for a non-technical stakeholder, starting from the decision they need to make.
• Write and say aloud your answer on what motivates you about industrial gas and energy work, grounded in a specific experience.

Deliverable: A stakeholder explanation, a sector-motivation answer, one finished practice exercise and notes from a full mock run.

Behavioural preparation

The reported behavioral questions cover technical disagreement, explaining complex ideas to non-technical people, pivoting on changing requirements, designing from scratch and motivation for the sector. For this role the stories should show you working with mechanical and electrical engineers, keeping safe failure behavior in view and making a clear decision yourself.

1. Describe a situation where you had to balance performance with safety requirements.

How do you handle real-time data processing in a resource-constrained environment?

Medium · Technical Proficiency and Domain Knowled

Approach

1. Open by pinning down the problem: what the data rate is, what the deadline is for each sample or batch, what happens if the deadline is missed (soft or hard), and what the memory and CPU budget actually is. Without those numbers nothing else is justified.
2. Describe the data path: an interrupt or DMA transfer fills a fixed-size ring buffer, the interrupt does the minimum (store, timestamp, set a flag), and a task or main loop does the processing. State the memory rule: allocate everything at start-up, none inside the processing loop.
3. Keep per-sample work bounded: integer or fixed-point arithmetic when the target has no floating-point unit, incremental algorithms (running sums, small filters) instead of keeping whole windows, and lookup tables when they are cheaper than computing.
4. Say how you verified it: measure worst-case execution time and worst-case queue depth on the target, not the average on a laptop. Give the overload policy (drop oldest or newest, counter for dropped samples) and a watchdog or error flag for the case where processing falls behind.
5. If you tell it as a story, use STAR with one measured result, for example how much headroom you recovered or how many samples were dropped before and after the change.

Follow-up

• What happens when the producer outruns the consumer, and who decides which samples are dropped?
• How did you measure worst-case latency, and what would change on a faster or slower processor?
• Why a ring buffer here instead of a queue that allocates per message?

Walk me through a time you had to design a system from scratch. What were your primary constraints?

Medium · Problem-Solving and System Design

Approach

1. Choose a system you owned from a blank page, and open with one sentence on what it had to do and for whom. Keep the situation short so most of the answer is about decisions.
2. Sort the constraints into groups and name the real ones: hardware limits (memory, processor, I/O), failure behavior (what the system must do when a part fails), interfaces to other teams or equipment, schedule, and what the team could maintain afterwards. Say which of them forced the architecture.
3. For the two or three biggest decisions, name the alternatives you rejected and the reason. Showing a trade-off (for example a simpler design that was easier to test versus a more flexible one) matters more than the final diagram.
4. Explain how you checked the design before committing: a prototype, a bench test, a review with the people who owned the hardware, or a failure walk-through. Close with the outcome and one thing you would change.
5. Avoid the mistake of listing features and technologies. A story with no stated constraint and no 'I decided' reads as a project description.

Follow-up

• Which constraint did you discover late, and what did it change?
• What alternative did you reject, and would you reject it again?
• How did you split the design into parts that could be tested independently?

Tell me about a time you had a technical disagreement with a colleague. How did you resolve it?

Medium · Behavioral and Collaboration

Approach

1. Pick a disagreement about a technical matter (an interface, a data structure, a testing approach, a language or a buffer strategy), not a personality clash. Name both positions fairly, including the strongest argument on the other side.
2. Show how you moved from opinion to evidence: agree on the criteria first (latency, memory, safety behavior, maintainability), then use a benchmark, a prototype, a failure case or a review with a third person to compare.
3. State the resolution and your part in it. Strong stories include the case where you changed your mind, or where you both adopted a hybrid, and say what that showed you.
4. Finish with the relationship and the follow-through: how the decision was written down so it would not be reopened, and how you worked together afterwards.
5. The mistake that sinks this answer is a story where you were simply right, the colleague was wrong and the matter ended by seniority.

Follow-up

• What evidence actually changed the outcome, and who gathered it?
• What would you have done if the colleague still disagreed after that?
• Looking back, were you right? What did you learn about when to push?

How do you explain complex technical concepts to non-technical stakeholders or clients?

Medium · Behavioral and Collaboration

Approach

1. Choose a real occasion: a mechanical engineer, a customer, a manager or a field technician who needed a decision from you. Start with what they needed to decide, because a good explanation is built backwards from that decision.
2. Replace jargon with their vocabulary and an analogy from their world (for a mechanical engineer, a feedback loop is a thermostat or a valve that overshoots). Translate technical limits into things they care about: downtime, safety margin, schedule, cost.
3. Describe how you made it concrete: a diagram, a short demo, a table of options with consequences, or a one-page summary sent afterwards.
4. Say how you checked they understood, by asking them to restate the trade-off or by seeing the decision they made, and what you changed when they did not.
5. The usual mistake is describing a talk you gave rather than a result. End on the decision or action that came from your explanation.

Follow-up

• How do you handle a stakeholder who wants a date you believe is unsafe or unrealistic?
• How did you know the explanation worked rather than just being nodded at?
• What do you leave out of an explanation, and how do you decide?

Describe a project where you had to pivot your strategy due to changing requirements.

Medium · Behavioral and Collaboration

Approach

1. Pick a change that was real: a requirement, a hardware revision or a priority that invalidated work already done. Say what you had built or planned and what changed, in two sentences.
2. Show the decision process: how you found out, what you estimated the cost of switching to be, what could be kept (interfaces, tests, modules) and what had to be thrown away. Naming sunk cost directly shows judgment.
3. Describe how you communicated the change: what you told the stakeholders about scope, schedule and risk, and what you asked them to trade away.
4. Explain how you protected quality while moving fast, with tests carried over, a short list of new risks and an early check with the people affected.
5. Close with the result and what you changed in your habits afterwards, for example isolating hardware-specific code behind an abstraction layer so a later change touches less. Avoid blaming the people who changed the requirement.

Follow-up

• How did you decide what to keep and what to discard?
• How did you explain the schedule impact, and what did you ask to be cut?
• What did you change in your design habits so the next change costs less?

What motivates you to work in the industrial gas and energy sector?

Medium · Behavioral and Collaboration

Approach

1. Answer with evidence rather than slogans. Name a concrete experience that pulled you toward software that controls physical equipment: a project with sensors or actuators, a course in control systems, or a time you fixed a fault that only existed on real hardware.
2. Show you understand the kind of work. Candidates report control software for gas delivery equipment, close collaboration with mechanical and electrical engineers, and occasional field support, so say which parts of that you want and why, including the less glamorous ones such as documentation and reviews.
3. Connect safety and reliability to your own habits: how you handle failure states, test, and document, using one small example.
4. Research the product category (what gas delivery equipment does at a high level) so you can speak accurately, and do not claim facts about the company you cannot back up.
5. Avoid generic mission statements and avoid saying you want any engineering job in the sector.

Follow-up

• What do you already know about the equipment your software would control?
• Which part of this work would you find least appealing, and how would you handle it?
• How does your experience with hardware or physical systems show up in the code you write?