Skip to content

Prep plans & questionnaires#

Analysis tells you where you stand. These two turn it into work.

Prep plans#

Build my prep plan produces a set of actionables for that specific role, deliberately composed rather than a uniform list:

Kind What it is
Coding Real problems, chosen for the patterns this company's stack and domain make likely. Each names its DS&A pattern.
Concept Targeted at the gaps the analysis found — not your strengths.
System design Framed as a prompt the interviewer might actually pose for this product, not a textbook topic.
Behavioral The specific story to prepare, not "think of a time when".
Research The company, its product, its engineering writing.

Every item carries a rationale explaining why it matters for this posting. That's the part worth reading — it's what separates a prep plan from a generic list, and it tells you what to drop when you're short on time.

A generated prep plan — coding problems with their pattern, concepts targeting the gaps, and a system-design prompt.

Items have a difficulty, an estimated duration, and a priority. The plan is sized for one to two weeks, not a month.

Working through it#

Click the circle to cycle an item: todo → in progress → done. The card header tracks completion and remaining hours.

Regenerate replaces untouched AI suggestions but keeps anything you've started, finished or added yourself.

The Prep page collects actionables across every application, filtered by kind, so you can work the queue rather than the board.

Doing the work in Alfred#

Coding, concept, system design and behavioural items are not checkboxes. Each opens a workspace where you do the thing, and Alfred decides whether it is done.

Solve it opens a coding problem: a brief, worked examples, a Python editor and a test suite.

  • Run samples runs the visible cases. Fast, repeatable, and the normal way to work — the equivalent of running it locally.
  • Submit runs every case, including the hidden ones, and is the only thing that can complete the item.

Hidden tests stay hidden, with one exception: if a hidden case fails, the first one is reported in full. A suite that only says "hidden test 9 failed" sends you guessing, and one that prints every expected value has no hidden tests at all.

Answer it opens a written prompt — a design question posed for the company you applied to, a concept to explain, a story to tell — with a rubric behind it. Submitting grades the answer against each requirement and tells you which ones you met. The rubric is available before you start, folded away, because reading it first turns the exercise into filling in a form.

Hints are there for all of them, revealed one at a time rather than all at once.

Why you can't just click done#

An item with an exercise behind it cannot be marked done. The API refuses it, so pressing the button harder will not help. done is set when a submission passes, and it records which one.

You can always skip an item. Declining a problem is honest; claiming you solved it is not.

Generated, and checked before you see it

The problems are written for your role rather than linked from a problem set — so they can carry a test suite, which is what makes checking your work possible.

Before a coding exercise is offered, Alfred solves it itself and runs its own solution against its own tests. If that fails, it tries once more; if it still fails, you are told the tests are not trustworthy and that item can be completed by hand. You are never held to a check Alfred got wrong.

Where your code runs#

On the server, in a WebAssembly sandbox, in a separate short-lived process with no filesystem or network access. Nothing the browser reports about whether you passed is trusted — a verdict the page can assert would be worth nothing.

A submission that loops forever is killed, and you are told which case never came back.

Questionnaires#

Draft the questions predicts the interview questions and drafts answers you can rehearse. Each has:

  • The question, as an interviewer would ask it.
  • What they're really asking — the thing being assessed underneath.
  • Alfred's draft — an answer in your voice, built from your stated background, STAR-structured for behavioural questions, under ~150 words because it's a rehearsal script rather than an essay.
  • Your version — the box that matters. Rewriting the draft in your own words is what makes it stick.
  • A readiness rating, so you can see what still needs work.

A predicted question, what the interviewer is really assessing, and a drafted answer to rewrite in your own words.

Questions are ordered by likelihood combined with how badly a weak answer would hurt, and at least two always target the gaps from the analysis — those are the ones that sink loops.

Regenerating keeps every question you've already answered.

Ask Alfred#

The application page has a chat with the posting and the analysis already in context. Useful for the things that don't fit a prep item: what to ask at the end, how to explain a gap, drafting a follow-up email, what to anchor on in a compensation conversation.