Where it comes from

How to answer

One candidate reported on Aced (formerly Exponent) that their Scale AI onsite, with the interview listed as August 2025, included a round debugging a giant repo. Source 1Scale AI Forward Deployed Engineer Interview Experience (2026)PublisherAced (formerly Exponent)Source typecandidate’s personal write-up This question practices one skill a round like that plausibly draws on: build an accurate map fast, from one request, without reading the whole codebase.

  1. Pin one concrete request. Method, path and an example body: POST /api/v2/invoices/<id>/send. Say the deliverable: every hop from the network edge to each store and external call, with a file and line, each marked ran, read or guessed.
  2. Start where the process starts. The Procfile, Dockerfile CMD or package.json scripts name the web entry point and any workers, and a worker runs the part the handler can’t show. Check for existing tracing (OpenTelemetry, an APM , request-id middleware): one real trace is a confirmed skeleton.
  3. Find the route the framework’s way. Ask for its route table (flask routes, bin/rails routes) or search a distinctive segment, not the full URL, because prefixes are attached elsewhere. If the string appears nowhere, the path may be rewritten before the app sees it: check the ingress, gateway or proxy config.
  4. Walk down, noting boundaries. Handler, service, data access; at each hop, note what crosses a boundary: database, cache, queue, HTTP call. List middleware in order without reading every line.
  5. Hunt the indirection search misses. Decorators, dependency injection, ORM hooks, feature flags, queue producers. For a queue, search the task name to find the consumer.
  6. Confirm with the runtime. A stack trace from the deepest function, one request id followed through both logs, or an existing test run with a print. Upgrade edges from read or guessed to ran.

Narrate as you go. The trap is reading files in folder order and presenting a guess as a map.

GlossaryForward deployed engineerA software engineer who builds and ships production systems inside a customer’s problem and environment, accountable to that customer’s outcome.More on Forward deployed engineerGlossaryAgentA system in which a model chooses steps and tool calls to complete a task, within limits the design sets.More on Agent

Follow-ups

What the interviewer may ask next, once your first answer is on the table.

  • The route string appears nowhere in the repo. How do you find the handler?
  • Your map says the email is sent, but the customer never got it. Which hop do you check first, and what would you look at there?
  • The path crosses into a service whose code you cannot open. What goes on the map for that hop?
  • You are nearly out of time. Which edges do you leave marked as inferred, and which must be confirmed?

Where answers go wrong

  • Reading outward from the README or the directory tree instead of from one concrete request, and ending the round with a vague picture of everything.
  • Drawing only the synchronous call chain, and missing the queue hop, the ORM hook or the before-request middleware that decides what actually happens.
  • Presenting edges guessed from naming conventions as fact, with nothing run to confirm them.

Answer this in two minutes

Write the answer you would say out loud. The clock starts with your first word.

Two minutes

Model answer

“The request is POST /api/v2/invoices/<id>/send. I’ll write the path down as I go, one line per hop, and mark each edge by its evidence: ran means I saw it at runtime, read means I read the code that does it, guessed means I inferred it from names or config I can’t see.

Learn this in Module 6: Screens and unfamiliar code, part of the curriculum.