Reducto Software Engineer Interview Guide

The 60-second brief

As a Software Engineer (specifically focusing on the Senior Software Engineer, Platform track) at Reducto, you will occupy a critical role at the intersection of high-scale systems engineering and applied artificial intelligence. Reducto is solving one of the most pervasive bottlenecks in the modern AI ecosystem: converting unstructured enterprise data—such as complex PDFs, financial statements, medical records, and multi-tab spreadsheets—into highly accurate, structured formats that LLMs and AI agents can actually utilize. Because nearly 80% of enterprise data is locked in these unstructured formats, your work directly determines how effectively leading AI teams and Fortune 500 enterprises can leverage their core data assets.

This guide is scoped to a Software Engineer candidate at Reducto.

Focus areas

• Python
• LLM integration into production systems
• LLM calls optimization

The Reducto interview process

Reducto candidates report 3 rounds over 3-5 weeks. The stages below are what candidates describe, not a published process.

Initial Conversation

A discussion with one of the founders or engineering leaders focusing on your background and alignment with the company's working style.

What to demonstrate

• A discussion with one of the founders or engineering leaders focusing on your background and alignment with the company's working style
• Depth in Python

How to prepare

• Answer aloud and timed: How do you design a reliable schema extraction pipeline using LLMs when the input document layout is highly variable and non-standard?
• Answer aloud and timed: What strategies do you employ to guarantee that an LLM output strictly conforms to a specific JSON schema in a production environment?

Technical Assessments

A series of assessments evaluating hands-on coding ability, system design skills, and problem-solving in document processing.

What to demonstrate

• A series of assessments evaluating hands-on coding ability, system design skills, and problem-solving in document processing
• Depth in Python

How to prepare

• Answer aloud and timed: Explain how you would design an evaluation framework to measure the accuracy and regression of prompt changes across thousands of diverse test documents.
• Answer aloud and timed: How do you handle context window limitations and token cost optimization when processing extremely long, text-heavy financial spreadsheets?

Onsite Interview

An intensive collaborative interview at the San Francisco office focusing on practical engineering problems and past technical achievements.

What to demonstrate

• An intensive collaborative interview at the San Francisco office focusing on practical engineering problems and past technical achievements
• Depth in Python

How to prepare

• Answer aloud and timed: Describe a time you had to debug a silent failure in an LLM-based extraction pipeline. How did you identify the root cause, and how did you prevent it from recurring?
• Answer aloud and timed: Design an asynchronous document processing pipeline that can ingest, OCR, chunk, and extract data from 10,000 multi-page PDFs simultaneously.

Where candidates lose points

Highlight Your Tool-Building Experience

Reducto highly values engineers who build their own tools to solve problems. If you have ever built a custom CLI tool, a Streamlit dashboard, or a debugging utility to make your own life easier, make sure to talk about it during your interviews.

Demonstrate High Agency

When describing past projects, focus on how you identified a problem yourself, took ownership of the solution, and shipped it. Avoid talking about projects where you simply executed a pre-written specification.

Be Quantitative

When explaining your past achievements, use concrete metrics. Instead of saying "I optimized a pipeline," say "I reduced backend latency by 35% and cut LLM token costs by 20% by implementing a custom semantic chunking algorithm."

Questions & practice

Given a list of bounding boxes representing text blocks on a page, write an algorithm to group them into logic

Medium · Algorithmic Problem Solving & Document P

Given a list of bounding boxes representing text blocks on a page, write an algorithm to group them into logical reading orders (columns and paragraphs).

Approach

1. Restate the input: its shape, its size, and what is guaranteed about it.
2. Name the brute-force solution and its complexity before improving on it.
3. Choose the data structure from the access pattern, not from familiarity.
4. State the target complexity and say which constraint rules the naive version out.

Follow-up

• How does this change if the input no longer fits in memory?
• What is the worst case, and how likely is it on real data?

How would you implement an efficient layout-aware text chunking algorithm that ensures chunks do not break in

Medium · Algorithmic Problem Solving & Document P

How would you implement an efficient layout-aware text chunking algorithm that ensures chunks do not break in the middle of a sentence or a markdown table row?

Approach

1. Restate the input: its shape, its size, and what is guaranteed about it.
2. Name the brute-force solution and its complexity before improving on it.
3. Choose the data structure from the access pattern, not from familiarity.
4. State the target complexity and say which constraint rules the naive version out.

Follow-up

• How does this change if the input no longer fits in memory?
• What is the worst case, and how likely is it on real data?

Write a parser to convert a raw, unstructured CSV representation of a financial balance sheet into a nested JS

Medium · Algorithmic Problem Solving & Document P

Write a parser to convert a raw, unstructured CSV representation of a financial balance sheet into a nested JSON structure representing parent-child account relationships.

Approach

1. Restate the input: its shape, its size, and what is guaranteed about it.
2. Name the brute-force solution and its complexity before improving on it.
3. Choose the data structure from the access pattern, not from familiarity.
4. State the target complexity and say which constraint rules the naive version out.

Follow-up

• How does this change if the input no longer fits in memory?
• What is the worst case, and how likely is it on real data?

Describe how you would optimize a CPU-bound PDF text extraction script in Python that is currently bottlenecke

Medium · Algorithmic Problem Solving & Document P

Describe how you would optimize a CPU-bound PDF text extraction script in Python that is currently bottlenecked by single-threaded execution.

Approach

1. Say what the runtime actually does before reasoning about the code.
2. Name what is shared across threads and what owns each piece of state.
3. Identify the window where an invariant is briefly untrue.
4. Distinguish a value from a reference to it, and say which one you handed out.

Follow-up

• What happens if two callers reach this at the same time?
• Where could this allocate more than you expect?

Describe a situation where you had to build a custom tool (such as a dashboard or utility script) to unblock y

Medium · Behavioral & Startup Alignment

Describe a situation where you had to build a custom tool (such as a dashboard or utility script) to unblock yourself or your team. What did you build, and what was the impact?

Approach

1. Say what the runtime actually does before reasoning about the code.
2. Name what is shared across threads and what owns each piece of state.
3. Identify the window where an invariant is briefly untrue.
4. Distinguish a value from a reference to it, and say which one you handed out.

Follow-up

• What happens if two callers reach this at the same time?
• Where could this allocate more than you expect?

Denormalise tenant onto revisions and backfill it live

Hard · denormalisation · online migration · backfill

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

Approach

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

Follow-up

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

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?

How do you design a reliable schema extraction pipeline using LLMs when the input document layout is highly va

Medium · LLM Orchestration & Structured Extractio

How do you design a reliable schema extraction pipeline using LLMs when the input document layout is highly variable and non-standard?

Approach

1. Fix the scope first: who calls this, how often, and what they do when it fails.
2. Name the read and write paths separately; they rarely have the same bottleneck.
3. Choose a partition key and say what query it makes expensive.
4. State the consistency you need, and where you are willing to be stale.

Follow-up

• What breaks first when traffic grows ten times?
• How does this behave when that dependency is down for an hour?

What strategies do you employ to guarantee that an LLM output strictly conforms to a specific JSON schema in a

Medium · LLM Orchestration & Structured Extractio

What strategies do you employ to guarantee that an LLM output strictly conforms to a specific JSON schema in a production environment?

Approach

1. Clarify what is being asked and what a complete answer contains.
2. State your assumptions explicitly before working the problem.
3. Say what you would check first and why it is the highest-information step.
4. Work from the requirement backwards to the design.

Follow-up

• What assumption would you test first?
• How would you know your answer was wrong?

Explain how you would design an evaluation framework to measure the accuracy and regression of prompt changes

Medium · LLM Orchestration & Structured Extractio

Explain how you would design an evaluation framework to measure the accuracy and regression of prompt changes across thousands of diverse test documents.

Approach

1. Fix the scope first: who calls this, how often, and what they do when it fails.
2. Name the read and write paths separately; they rarely have the same bottleneck.
3. Choose a partition key and say what query it makes expensive.
4. State the consistency you need, and where you are willing to be stale.

Follow-up

• What breaks first when traffic grows ten times?
• How does this behave when that dependency is down for an hour?

Design an asynchronous document processing pipeline that can ingest, OCR, chunk, and extract data from 10,000

Medium · System Design & Pipeline Architecture

Design an asynchronous document processing pipeline that can ingest, OCR, chunk, and extract data from 10,000 multi-page PDFs simultaneously.

Approach

1. Fix the scope first: who calls this, how often, and what they do when it fails.
2. Name the read and write paths separately; they rarely have the same bottleneck.
3. Choose a partition key and say what query it makes expensive.
4. State the consistency you need, and where you are willing to be stale.

Follow-up

• What breaks first when traffic grows ten times?
• How does this behave when that dependency is down for an hour?

How would you architecture a caching layer for document layout analysis to avoid redundant ML model inference

Medium · System Design & Pipeline Architecture

How would you architecture a caching layer for document layout analysis to avoid redundant ML model inference and minimize processing costs?

Approach

1. Fix the scope first: who calls this, how often, and what they do when it fails.
2. Name the read and write paths separately; they rarely have the same bottleneck.
3. Choose a partition key and say what query it makes expensive.
4. State the consistency you need, and where you are willing to be stale.

Follow-up

• What breaks first when traffic grows ten times?
• How does this behave when that dependency is down for an hour?

Describe how you would design a rate-limiting and fair-use queuing system for a multi-tenant API where some cu

Medium · System Design & Pipeline Architecture

Describe how you would design a rate-limiting and fair-use queuing system for a multi-tenant API where some customers submit single-page documents and others submit 1,000-page documents.

Approach

1. Fix the scope first: who calls this, how often, and what they do when it fails.
2. Name the read and write paths separately; they rarely have the same bottleneck.
3. Choose a partition key and say what query it makes expensive.
4. State the consistency you need, and where you are willing to be stale.

Follow-up

• What breaks first when traffic grows ten times?
• How does this behave when that dependency is down for an hour?

What architectural trade-offs do you consider when deciding whether to run document pre-processing (like image

Medium · System Design & Pipeline Architecture

What architectural trade-offs do you consider when deciding whether to run document pre-processing (like image rotation or table detection) on CPU vs. GPU instances?

Approach

1. Fix the scope first: who calls this, how often, and what they do when it fails.
2. Name the read and write paths separately; they rarely have the same bottleneck.
3. Choose a partition key and say what query it makes expensive.
4. State the consistency you need, and where you are willing to be stale.

Follow-up

• What breaks first when traffic grows ten times?
• How does this behave when that dependency is down for an hour?

Describe a time you had to debug a silent failure in an LLM-based extraction pipeline. How did you identify th

Medium · LLM Orchestration & Structured Extractio

Describe a time you had to debug a silent failure in an LLM-based extraction pipeline. How did you identify the root cause, and how did you prevent it from recurring?

Approach

1. Establish what changed and when, before forming any theory.
2. Pick a bisection that eliminates candidates whichever way it turns out.
3. Check the instrumentation before believing the symptom.
4. Separate the trigger from the cause; the deploy is rarely the bug.

Follow-up

• What would you look at first, and what would it rule out?
• How would you tell a cause from a coincidence here?

Suggested practice checklist

Built from the rounds and topics Reducto candidates report.

Day 1 — Map the Reducto loop

• Write out the reported sequence: Initial Conversation, Technical Assessments, Onsite Interview.
• For each round, write one sentence on what it is judging, from the description above, and mark the one you are least ready for.

Deliverable: A one-page map of the 3 reported rounds, with the weakest marked.

Day 2 — Work Python

• Spend the session on Python, which Reducto candidates report being tested on.
• Write one worked example in Python and time yourself on it.

Deliverable: One timed worked example in Python.

Day 3 — Work LLM integration into production systems

• Spend the session on LLM integration into production systems, which Reducto candidates report being tested on.
• Write one worked example in LLM integration into production systems and time yourself on it.

Deliverable: One timed worked example in LLM integration into production systems.

Day 4 — Work LLM calls optimization

• Spend the session on LLM calls optimization, which Reducto candidates report being tested on.
• Write one worked example in LLM calls optimization and time yourself on it.

Deliverable: One timed worked example in LLM calls optimization.

Day 5 — Answer out loud: LLM Orchestration & Structured Extraction

• Answer aloud, timed: How do you design a reliable schema extraction pipeline using LLMs when the input document layout is highly variable and non-standard?
• Answer aloud, timed: What strategies do you employ to guarantee that an LLM output strictly conforms to a specific JSON schema in a production environment?

Deliverable: Spoken answers to 2 reported LLM Orchestration & Structured Extraction question(s), under time.

Day 6 — Answer out loud: System Design & Pipeline Architecture

• Answer aloud, timed: Design an asynchronous document processing pipeline that can ingest, OCR, chunk, and extract data from 10,000 multi-page PDFs simultaneously.
• Answer aloud, timed: How would you architecture a caching layer for document layout analysis to avoid redundant ML model inference and minimize processing costs?

Deliverable: Spoken answers to 2 reported System Design & Pipeline Architecture question(s), under time.

Day 7 — Answer out loud: Algorithmic Problem Solving & Document Parsing

• Answer aloud, timed: Given a list of bounding boxes representing text blocks on a page, write an algorithm to group them into logical reading orders (columns and paragraphs).
• Answer aloud, timed: How would you implement an efficient layout-aware text chunking algorithm that ensures chunks do not break in the middle of a sentence or a markdown table row?

Deliverable: Spoken answers to 2 reported Algorithmic Problem Solving & Document Parsing question(s), under time.

Behavioural preparation

Behavioural rounds judge the decision you made and what it cost.

How do you handle context window limitations and token cost optimization when processing extremely long, text-

Medium · LLM Orchestration & Structured Extractio

How do you handle context window limitations and token cost optimization when processing extremely long, text-heavy financial spreadsheets?

Approach

1. Pick a story where you made the decision, not one where you watched it.
2. State the situation in two sentences and spend the rest on the reasoning.
3. Give the blast radius: what could have broken, and what you measured.
4. Name the disagreement and how you resolved it with evidence.

Follow-up

• What would you do differently if you ran that again?
• How did you know your change caused the improvement?

How do you handle state management and recovery in a multi-stage document parsing pipeline if a worker node cr

Medium · System Design & Pipeline Architecture

How do you handle state management and recovery in a multi-stage document parsing pipeline if a worker node crashes mid-process?

Approach

1. Pick a story where you made the decision, not one where you watched it.
2. State the situation in two sentences and spend the rest on the reasoning.
3. Give the blast radius: what could have broken, and what you measured.
4. Name the disagreement and how you resolved it with evidence.

Follow-up

• What would you do differently if you ran that again?
• How did you know your change caused the improvement?

Talk about a project where you took a feature from complete ambiguity to production shipping in a matter of da

Medium · Behavioral & Startup Alignment

Talk about a project where you took a feature from complete ambiguity to production shipping in a matter of days. What trade-offs did you make?

Approach

1. Pick a story where you made the decision, not one where you watched it.
2. State the situation in two sentences and spend the rest on the reasoning.
3. Give the blast radius: what could have broken, and what you measured.
4. Name the disagreement and how you resolved it with evidence.

Follow-up

• What would you do differently if you ran that again?
• How did you know your change caused the improvement?

Reducto emphasizes being "your own worst critic." Give an example of a time you shipped something that met the

Medium · Behavioral & Startup Alignment

Reducto emphasizes being "your own worst critic." Give an example of a time you shipped something that met the initial requirements but you chose to rebuild or heavily refactor because it didn't meet your personal bar for quality.

Approach

1. Pick a story where you made the decision, not one where you watched it.
2. State the situation in two sentences and spend the rest on the reasoning.
3. Give the blast radius: what could have broken, and what you measured.
4. Name the disagreement and how you resolved it with evidence.

Follow-up

• What would you do differently if you ran that again?
• How did you know your change caused the improvement?

How do you prioritize engineering tasks when you have competing demands from founders, high-value enterprise c

Medium · Behavioral & Startup Alignment

How do you prioritize engineering tasks when you have competing demands from founders, high-value enterprise customers, and technical debt?

Approach

1. Pick a story where you made the decision, not one where you watched it.
2. State the situation in two sentences and spend the rest on the reasoning.
3. Give the blast radius: what could have broken, and what you measured.
4. Name the disagreement and how you resolved it with evidence.

Follow-up

• What would you do differently if you ran that again?
• How did you know your change caused the improvement?