Axsys IT Software Engineer Interview Guide

The 60-second brief

Candidates describe Axsys IT as having two product areas that engineers work in: Embedded Finance platforms and a Data & AI division. Software Engineers build APIs for the Embedded Finance products and work on data pipelines for the Data & AI side. The role covers the whole development lifecycle of a feature, from turning product requirements into a technical design, through code review, to keeping production healthy. Reported nice-to-have experience includes fintech or AI/ML work, cloud-native infrastructure and automated CI/CD pipelines.

This guide is for candidates interviewing for a Software Engineer position at Axsys IT. It covers the five questions candidates report (state management, database query tuning, production debugging, API architecture trade-offs in Embedded Finance, and security for financial data), the topic areas that come up (full-stack development, Embedded Finance systems, Data & AI integration, general software engineering and frontend) and a seven-day plan to rehearse them.

No fixed round order is reported for Axsys IT. Candidates describe topics and questions instead, so confirm the stages and format with your recruiter and prepare across all of them.

Focus areas

• Full Stack Development
• Embedded Finance Systems
• Data & AI Integration

Overview

Candidates describe the role as full-stack work with two flavours. On the Embedded Finance side it is API design and the services behind those APIs. On the Data & AI side it is data pipelines and getting models into production. The requirements candidates list are full-stack development, experience with distributed systems and solid database design, so expect to be asked about all three layers: what the browser holds, what the service does, and how the data is stored and queried.

The five reported questions are all open-ended explanations rather than puzzles. Each one is answered best with a concrete system from your own experience: the state you managed, the slow query you fixed, the incident you debugged, the API you shaped, the sensitive data you protected. Prepare one real example per question and a general framework behind it, so you can switch between the story and the principle when asked to go deeper.

Because the reported material includes Data & AI integration (ETL, model deployment, latency and accuracy monitoring in production) as well as Embedded Finance, find out early which team you are interviewing for and weight your preparation accordingly. Candidates also report that explaining technical decisions to non-technical colleagues matters, and that STAR-structured answers keep stories focused on your own contribution.

Length of preparation should go into depth on trade-offs: why one API style over another, why one index over another, why a given consistency or caching choice. Write these out and say them aloud, since the questions reward a structured spoken answer more than recall of a fact.

Where to focus your preparation

Candidates do not report a fixed sequence of rounds for Axsys IT, so the loop is shown as one preparation focus covering the reported topics.

Preparation focus

No round sequence has been reported, only the topics and questions candidates meet. These are full-stack development, Embedded Finance systems, Data & AI integration, general software engineering and frontend development, and the reported questions are explanation and design questions rather than a named coding exercise. Ask your recruiter how many stages there are, who you will speak with, whether you will write code live, and which team (Embedded Finance or Data & AI) the role belongs to. Then prepare each reported question as a structured spoken answer with one real example behind it.

What to demonstrate

• Whether you can explain full-stack decisions such as state handling and data flow from database to client
• Whether you reason about database performance and API trade-offs with specifics rather than slogans
• Whether you can describe a real production incident and your debugging method clearly

How to prepare

• Write a one-page answer outline for each of the five reported questions, with one example from your own work attached
• Practise the state management and query tuning answers aloud until each stands as a structured explanation without notes
• Prepare a STAR story for the production debugging question with the symptom, your steps, the cause and the fix
• Send the recruiter your questions about format, coding expectations and which team the role sits on

Where candidates lose points

Answering the state management question by naming a library

Separate server cache state, client UI state, URL state and session state, say where each lives and how it is invalidated, then cover optimistic updates, rollback on failure and stale reads under heavy traffic. The library is a detail after the model.

Answering the query optimization question with only 'add an index'

Describe the sequence: find the slow statement, read the plan with estimated against actual row counts, check for sequential scans, sorts that spill and bad join order, then choose composite index column order, keyset pagination, narrower selects or partitioning. Mention the write cost of every new index.

Comparing API styles in the abstract without the money-moving context

For the Embedded Finance question, tie each style (REST, GraphQL, gRPC, webhooks or events) to idempotency keys, retries, versioning for partner integrations, authentication scopes, rate limits and auditability. Show what happens when a payment-style call times out and is retried.

Giving the security and compliance answer as a list of buzzwords

Structure it by data lifecycle: classify the data, encrypt in transit and at rest, tokenize or mask fields you do not need, apply least-privilege access, keep secrets out of code, write audit logs and set retention. Say you would confirm which regulations apply to the product rather than assuming them.

Telling the production debugging story without a method or numbers

Use STAR with a timeline: first signal, what you checked first and why, what you ruled out, how you proved the cause, the fix and the follow-up that prevents a repeat. State what you personally did and the measurable effect.

Questions & practice

Identify the heaviest tenants in a five-minute window under memory pressure

Medium · top-k · heavy hitters · streaming

The edge service handles about 3,000 requests per second across roughly 50,000 tenants, peaking near 9,000. Expose the 50 heaviest tenants by request count over the trailing five minutes so limits can be tightened before one tenant's backfill starves the fleet. You may not retain five minutes of raw records. Give the exact solution and its memory, then the bounded-memory approximation with its error stated as a formula, and say which you would ship and at what tenant cardinality that choice changes.

Approach

1. Do the exact version first, because it is affordable at this cardinality: a ring of 300 one-second counters per tenant, advanced lazily, is 1,200 bytes of counters per tenant and roughly 60 to 90 MB for 50,000 tenants with overhead. Carry a running total and subtract the bucket you overwrite so a window read is O(1) rather than 300 adds.
2. Extract the top 50 with a size-k min-heap over the tenant sums: O(d log k) for d tenants, against O(d log d) to sort them all. Maintaining the heap continuously instead of on query requires a tenant-to-heap-index map, because incrementing a count already inside the heap means sifting from a known position, and without that map you rebuild the heap on every request.
3. State the approximation precisely rather than gesturing at sketches. Misra-Gries with m counters retains every item whose true count exceeds N/(m+1), and each retained count underestimates the truth by at most N/(m+1). With m = 1,000 and N = 900,000 requests in the window the error is roughly 900 requests, which is fine for spotting a tenant sending 50,000 and useless for ranking two tenants 200 apart.
4. Say what breaks when the window slides: Misra-Gries and Space-Saving are insert-only and cannot be decremented as records age out. The workable construction is one summary per sub-window, say ten seconds, with 30 summaries merged at query time, and the merged error is the sum of the per-summary errors, so the bound degrades linearly in the number of sub-windows.

Follow-up

• The heaviest tenant is heavy because of one export job rather than user traffic. Should the limiter treat those as the same tenant?
• Two tenants sit tied at the boundary of the top 50. Does your answer flap, and does the flapping matter?

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?

Diff a projection against the primary without per-row point reads

Hard · reconciliation · range hashing · throttling

The listing projection has drifted and some rows show a stale version. The primary holds 40,000,000 resource rows across 12,000 tenants while serving 1,200 writes and 14,000 reads per second. The obvious repair, reading each resource row and comparing its version against the projection, is correct and would eventually finish. Explain precisely why it is unacceptable here, then give a diff that finds the differing rows, state its complexity, and make it safe to run against a live primary. Replication lag is usually under 100 ms and is not bounded.

Approach

1. Quantify the naive cost rather than calling it slow: 40,000,000 point reads at even 0.5 ms each is over five hours serialised, and the only lever is concurrency, which is exactly what you cannot spend. The primary's pool is sized for the write path, and 40,000,000 random reads evict the buffer cache that sustains the 85 percent cache hit rate, so the audit degrades the system it is auditing.
2. Replace random access with one ordered pass per side. Both sides can be read in (tenant_id, resource_id) order, which is a sequential scan on each and a merge join in O(n) time and O(1) memory. For a dense diff that is the whole answer, and it reads the primary once instead of 40,000,000 times.
3. For the expected sparse case, compare range hashes instead of rows: partition the key space, compute per range an order-independent aggregate over hash(resource_id, version), compare aggregates, and descend only into ranges that differ. With d differing rows and branching factor B, at most d ranges mismatch per level, so the drill-down examines O(d log_B(n/d)) ranges and reads full rows only in mismatching leaves.
4. Aggregate with a sum modulo 2^64 or a multiset hash, never XOR. XOR is order-independent but self-cancelling, so two rows wrong in the same way, or a row duplicated on one side, leave the range aggregate matching and the range is declared clean.

Follow-up

• The diff reports 900 stale rows. How do you decide between patching those rows and rebuilding the projection from resource_revision?
• Same job, but the projection lives in a search index that cannot be scanned in key order. What changes?

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?

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?

Explain how you manage state in a complex, high-traffic application.

Medium · Technical & Domain Expertise

Approach

1. Split the state into three kinds before naming any tool. UI state lives in the component or a small store. Server state is a cached copy of data the backend owns. Durable state is the database. Most high-traffic problems come from treating cached server state as if it were the source of truth.
2. Keep app servers stateless so a load balancer can send any request to any instance. Sessions go in a signed token (a JWT checked on every request) or in a shared store such as Redis with a TTL. Sticky sessions break when an instance dies and they unbalance load.
3. On the client, use a server-state cache (React Query or SWR style) with stale-while-revalidate and invalidation keyed by resource. Use a global store (Redux, Zustand) only for state that is truly cross-cutting, and say why prop-drilling or context does not fit.
4. For concurrent writes to shared state, use optimistic concurrency: add a version column, write UPDATE ... WHERE id = ? AND version = ?, and treat zero affected rows as a conflict to retry or surface. It is not a success. Pair it with idempotency keys so client retries do not apply the same change twice.
5. Cover the cache edge cases: a stampede when a hot key expires (use a single-flight lock or jittered TTLs), stale reads after a write (invalidate or write-through), and optimistic UI updates that need a rollback path when the server rejects them.

Follow-up

• Two users edit the same record at once. Walk through exactly what each of them sees.
• Your Redis session store goes down. What happens to logged-in users, and what would you change?
• When would you accept stale data in the UI, and how do you bound how stale it can get?

How do you approach optimizing database queries for large-scale data sets?

Medium · Technical & Domain Expertise

Approach

1. Measure before changing anything. Find the slow queries from the slow-query log or pg_stat_statements, ranked by total time (calls times mean) and not only by worst single run. Then read EXPLAIN (ANALYZE, BUFFERS) and compare estimated rows with actual rows at each node.
2. Fix access paths with the right index. A composite index serves equality columns first and then one range or sort column (leftmost-prefix rule). A covering index lets the planner use an index-only scan. Keep predicates sargable: WHERE created_at >= ? can use an index on created_at, but WHERE DATE(created_at) = ? cannot unless you add an expression index.
3. Fix the query shape. Remove N+1 loops by batching with IN or a join. Select only the columns you need. Replace deep OFFSET pagination, which still reads and discards every skipped row, with keyset pagination (WHERE (created_at, id) < (?, ?) ORDER BY created_at DESC, id DESC LIMIT n).
4. For very large tables, partition by time or tenant so queries prune partitions. Move heavy aggregates to a materialized view or a summary table refreshed on a schedule, and send read-only reporting traffic to a replica, accepting replication lag.
5. State the costs. Every index slows inserts and updates and takes space. Stale planner statistics give bad plans, so check when ANALYZE last ran. Verify a fix by rerunning EXPLAIN ANALYZE on production-like data volumes, not a small dev table.

Follow-up

• The planner ignores your new index. What are the likely reasons, and how do you check each one?
• How would you add an index to a large table that serves live writes without blocking them?
• When does denormalising beat adding another index?

What are the trade-offs between different API architectures in an Embedded Finance context?

Medium · Technical & Domain Expertise

Approach

1. Frame the choice by who calls the API. External partners embedding a financial product usually need REST over HTTPS with an OpenAPI spec and explicit versioning. Internal service-to-service calls can use gRPC: HTTP/2, protobuf contracts and streaming, but browsers need gRPC-Web or a gateway. GraphQL suits clients that need flexible reads, but HTTP caching is harder and authorization must hold per field.
2. Treat money-moving writes separately from reads. Every POST that creates a payment or transfer should accept an idempotency key, stored with the result, so a client retry after a timeout returns the original outcome instead of charging twice.
3. Many financial operations finish asynchronously (pending, then settled or failed), so pair the synchronous API with webhooks or an event stream. Webhooks need HMAC signatures, a timestamp to stop replays, retries with exponential backoff, and receivers that deduplicate, because delivery is at-least-once.
4. Compare coupling and failure modes. Synchronous chains add latency and spread outages, so use timeouts and circuit breakers. Event-driven designs decouple services but are only eventually consistent and need an outbox pattern so the database write and the published event cannot diverge.
5. Close with the operational side: rate limits per partner, versioning and deprecation of fields, consistent error codes partners can act on, and a full audit trail of requests and responses for reconciliation.

Follow-up

• A partner's payment request times out on their side. How does your design make sure it is not applied twice?
• How would you version this API when a response field has to change shape for existing integrators?
• Why might you choose polling over webhooks for some partners?

How do you ensure security and compliance when handling sensitive financial data?

Medium · Technical & Domain Expertise

Approach

1. Start with data classification. Identify which fields are sensitive (card numbers, bank account numbers, government IDs), where they flow, and who needs them. Then reduce scope: tokenise card data through a vault or processor so most services only ever see a token. That is the main way to shrink PCI DSS scope.
2. Encrypt in transit with TLS on every hop, internal ones included. Encrypt at rest with keys held in a KMS or HSM using envelope encryption: data keys encrypt the data and a master key encrypts the data keys. Rotate keys, and add field-level encryption for the most sensitive columns.
3. Control access with least privilege: role-based access, MFA for operators, short-lived credentials, and secrets in a secrets manager rather than in code or environment files. Log every access to sensitive records in an append-only audit trail.
4. Prevent leaks at the edges. Mask or redact sensitive fields in logs, traces and error messages. Use bind parameters to stop SQL injection, validate input at the API boundary, and keep production data out of test environments.
5. Tie it to compliance obligations. PCI DSS forbids storing sensitive authentication data such as the CVV after authorization. Retention rules can conflict with erasure requests under laws such as GDPR, which crypto-shredding (deleting the key) can resolve for data that also sits in backups. Regular access reviews and penetration tests are the evidence an audit asks for.

Follow-up

• An engineer needs production data to debug a payment bug. How do you give them what they need without exposing raw account numbers?
• How do you rotate an encryption key for a table with billions of rows without downtime?
• A user asks for deletion, but financial regulation requires keeping transaction records. How do you handle both?

Describe a time you had to debug a difficult production issue; what was your process?

Medium · Technical & Domain Expertise

Approach

1. Choose an incident where you personally drove the investigation and can name the mechanism you proved. Structure it with STAR, and spend most of the time on the Action: the order of your steps and why you took them in that order.
2. Open with the symptom as you first saw it (an error rate, a latency percentile, a customer report) and how you sized the blast radius: which users, endpoints or tenants, and since when. Say what you did to mitigate before you knew the cause, such as a rollback, a feature-flag off or shedding traffic.
3. Show a narrowing sequence. Line up the timeline against deploys, config and traffic changes. Use logs, metrics and traces to isolate one component. Form one hypothesis at a time, and choose checks that rule out half the candidates whichever way they turn out. Reproduce the failure outside production before calling it solved.
4. Separate trigger from root cause. For example, a deploy may have exposed a connection-pool limit or a race condition that already existed. Say how you confirmed the fix: the metric returned to baseline and a regression test failed before the fix and passed after it.
5. End with the prevention work you shipped (an alert, a test, a guardrail, a runbook) and a number: minutes to mitigate, minutes to resolve, users affected. A story that ends with 'it went away after a restart' and no proven cause is weak.

Follow-up

• What was the first hypothesis you ruled out, and what evidence ruled it out?
• If the same symptom came back next month, how would you know quickly whether it was the same cause?
• What would you change about your team's monitoring or alerting because of this incident?

Suggested practice checklist

The plan starts by confirming the format, then works through the five reported questions grouped by theme, adds the Data & AI topic, and ends with a full spoken mock.

Day 1 — Confirm the format and map the topics

• Ask the recruiter for the stages, whether you will write code live, and whether the role is on the Embedded Finance or Data & AI side
• Write a topic map of the five reported areas (full-stack, Embedded Finance systems, Data & AI integration, general engineering, frontend) and rate yourself on each
• Match one project from your own work to each stated requirement: full-stack, distributed systems, database design

Deliverable: A sent recruiter message, plus a topic map with a self-rating and a project for each requirement.

Day 2 — State management under heavy traffic

• Pick an application you built and annotate where server cache, client UI, URL and session state each live
• Outline how you handle invalidation, optimistic updates, rollback and stale reads when traffic is high
• Say the answer aloud as a structured explanation, then once more with your real example

Deliverable: A one-page state map of a real application and a recorded or spoken answer to the state management question.

Day 3 — Database query tuning

• Write your tuning routine as an ordered checklist: identify the statement, read the plan, compare estimated and actual rows, fix scans, joins and sorts, then measure again
• Work the SQL exercise about tag and share joins fanning out a page and note which join shape caused the duplication
• Work the SQL exercise about holding a per-tenant active cap against concurrent creates and note how you serialise the check

Deliverable: A written tuning checklist and two solved SQL exercises with a sentence on the trade-off in each.

Day 4 — API architecture and financial-data security

• Build a comparison table of REST, GraphQL, gRPC and webhooks or events for a money-moving integration, with rows for idempotency, versioning, partner authentication, rate limits and retries
• Write a data-lifecycle security outline: classification, encryption, tokenization or masking, least privilege, secrets, audit logging, retention
• List the questions you would ask to learn which compliance obligations apply to the product

Deliverable: One comparison table, one security outline and a list of clarifying questions.

Day 5 — Production debugging story

• Write a STAR story about a real production issue with a timeline from first signal to prevention
• Add the numbers you can state: error rate, users or tenants affected, time to mitigate, time to resolve

Deliverable: One STAR story with measurements, rehearsed aloud, plus notes from the outage-narration exercise.

Day 6 — Data & AI integration

• Sketch an ETL pipeline from source to warehouse covering ingestion, idempotent loads, schema changes and backfills
• Sketch a model serving path covering batch against online scoring, versioning, rollout and how you track latency and accuracy
• Work the coding exercises on collapsing a redelivered event batch and diffing a projection against the primary

Deliverable: Two diagrams (pipeline and serving path) and two solved coding exercises.

Day 7 — Full spoken mock and behavioral polish

• Answer all five reported questions aloud back to back, each with its example, and note where you wandered
• Rehearse the code review disagreement and the broken-integration exercises from the practice set
• Write the questions you want to ask interviewers about the team, the stack and how engineers work with product

Deliverable: A completed mock with weak spots marked and a written list of questions for interviewers.

Stories to prepare: debugging, ownership, cross-team work

Candidates report that the role involves working with product managers, joining code reviews and architecture discussions, and explaining technical decisions to non-technical colleagues. They also recommend STAR for structuring stories. Prepare stories that show what you decided, what it cost and what you measured, with your own contribution clear.

1. Describe a time you had to debug a difficult production issue; what was your process?
2. Tell me about a feature you owned from conception to deployment. What did you decide and what changed after launch?
3. Describe a time you explained a technical trade-off to a non-technical stakeholder. How did you know they understood it?
4. Tell me about technical debt you identified on your own. How did you make the case for fixing it?
5. Describe a code review disagreement and how it was resolved.

Tell callers you do not own that their integration breaks

Medium · deprecation · compatibility · stakeholders

A field in a write endpoint's response must change shape. You own the endpoint; you do not own the four internal callers or the outbound webhook consumers who read it. Describe a deprecation you were responsible for: what you shipped first, how you established who was actually reading the field, the window you gave and what set its length, what you did about the consumer who never moved, and how you decided removal was safe. Name the signal you used, not the announcement you sent.

Approach

1. Establish the reader set empirically rather than from a wiki of owners: per-field usage counters keyed by principal, or access logs attributed to a consumer. State the blind spot of whichever you pick, since a consumer that reads the field only on a monthly job will not appear in a week of logs.
2. Ship additive first. Populate the new field alongside the old one so no reader is forced to move, which is also what keeps a rolling deploy safe, because old and new instances answer the same requests at the same time and a rollback must still find the old shape present.
3. Set the window from the slowest legitimate consumer's release cadence, not from your calendar, and decide separately what to do for a consumer with no release process at all, such as an external webhook endpoint you can only email.
4. Convert silence into evidence before you rely on it: a short, low-traffic removal window that makes a still-dependent consumer fail visibly and loudly while you are watching, rather than at three in the morning after you have moved on.

Follow-up

• How would you detect a consumer that reads the field only during a monthly export?
• One caller refuses to move and has a commercial relationship behind it. What changes in your plan and what does not?

Narrate an outage you owned from page to postmortem

Hard · incident response · blast radius · postmortems

Pick an incident you personally drove, ideally one where writes were affected rather than reads. In six to eight minutes: state the symptom as it first appeared on a dashboard, the blast radius you established before you knew the cause, the mitigation you applied and when, the mechanism you eventually proved, and the follow-up that would prevent a repeat. Bring numbers: error rate, tenants affected, minutes to mitigate, minutes to resolve. If you cannot name what you measured, choose a different incident.

Approach

1. Open on the signal rather than the cause: which metric at which percentile moved, on which service, at what time, so the listener follows the same evidence you had rather than a conclusion you already reached.
2. Separate mitigation from diagnosis out loud. State what you did to stop the bleeding (flag off, shed traffic, drain a lease, roll back a deploy) and say plainly that you did it before the mechanism was known, because those are two jobs with different deadlines.
3. Establish blast radius in countable terms: how many tenants, how many writes, and crucially whether the effect was loss or only delay. An append-only revision table or a pending outbox row means the change survived and the projection was merely behind, which is a repair rather than a data-loss incident.
4. Prove the mechanism instead of asserting it. Name the trace span that grew, the plan that flipped to a sequential scan, the lease that expired, plus one alternative you ruled out and the signal that stayed flat while you ruled it out.

Follow-up

• What would you do differently in the first five minutes, given the same dashboard and no more information?
• Which follow-up action did you deliberately not take, and why was dropping it the right call?

Turn a code review disagreement into a decision

Easy · code review · optimistic concurrency · communication

A colleague's change updates a row with UPDATE resource SET version = version + 1 WHERE resource_id = $1 AND version = $2 and treats an affected-row count of zero as a successful no-op. You read that as a silently lost update; they think returning 200 is friendlier to clients than returning a conflict. Describe how you have handled a review disagreement of this shape: what goes in the comment, when you leave the thread, and who decides. Then write the comment you would leave here, in under 80 words.

Approach

1. Sort the disagreement before writing anything. A silently discarded write is a correctness claim about data; the choice between 409 and 412 is taste. Only the first justifies blocking a merge, and saying which one you are doing is most of the value of the comment.
2. Make the claim reproducible in the comment itself with an interleaving rather than a principle: A reads version 7, B reads version 7, B commits version 8, A's predicate matches zero rows, A is told it succeeded and A's edit is gone.
3. Offer the alternative with its cost attached: return 409 carrying the current version and the revision that won, so the client can re-read and re-apply. Note that automatic retry is not the fix, because a retry re-reads the winner's state and reapplies an intent formed against data that no longer exists.
4. Apply an escalation rule you can state: two round trips on the thread, then a call, and the service's owner decides rather than the reviewer. A reviewer who cannot be overruled is a bottleneck with extra steps.

Follow-up

• Where would you put the test that fails if someone reintroduces the swallowed zero rowcount?
• The author says clients cannot handle a 409. How do you check whether that is true?