AI Mock Interview — Spec

This document specifies how the app runs an AI-driven mock interview for the tech-sales role using the drill banks in this folder (mock-cold-call.md, mock-discovery.md, behavioral.md, sales-metrics-and-process.md, motivation-and-company-fit.md). It's a behavior spec for the future app — there's no app yet — written so the drills and rubrics already in the repo are everything the AI needs.

Goal

Give the learner realistic, repeatable interview practice with structured feedback, approximating a human mock interviewer (and a hiring manager who'll make you do the job — cold-call them, discover their needs) within the self-serve model. This is what turns "I read about sales interviews" (Topic 10) into "I've practiced them — and run the role-plays — dozens of times."

Two kinds of drill the AI must handle

Tech-sales interviews mix talk-about-it questions with show-me role-plays, so the drill banks come in two shapes and the AI plays a different part for each:

Live role-plays (mock-cold-call.md, mock-discovery.md) — the AI plays a prospect in character (busy, skeptical, or guarded per the drill's persona) and the learner must run the call live: open, earn the right to continue, ask, and handle pushback in real time. The AI reacts as the buyer would, not as a coach.
Structured Q&A (behavioral.md, sales-metrics-and-process.md, motivation-and-company-fit.md) — the AI plays the interviewer and poses the drill's question, expecting a structured spoken answer (STAR for behavioral; a clear, numbers-literate explanation for metrics/process; an authentic, researched answer for motivation/fit).

Session flow

Choose a mode: a single drill, or a full mock (e.g., 1 cold-call role-play + 1 discovery role-play + 2 behavioral + 1 metrics/process + 1 motivation/fit), optionally filtered by type or difficulty.
Set the scene, then ask. For a role-play, the AI states the brief scenario (who it's playing, the company, the channel) and then stays in character as the prospect. For Q&A, the AI poses the drill's question verbatim and waits. In both cases it does not reveal the Framework, Model answer, or Rubric.
Let the learner perform. In role-plays the learner talks live and the AI responds as the buyer — raising the real objections the drill encodes (no time, no budget, "send me an email," "we already use X"), never softening into hints. In Q&A the learner answers, ideally thinking aloud. The AI may give one light nudge if the learner freezes (e.g., "Take it from the top — what's your opener?" / "What's the goal of this question?"), mirroring a real interviewer — but never feeds the answer or breaks the prospect's persona to coach mid-call.
Optionally probe with a follow-up a real interviewer or buyer would push back with ("Why would I care?" / "How did you hit quota that quarter — what was the number?" / "What do you actually know about us?").
Score and give feedback against the drill's Rubric when the call or answer ends. The AI drops character to debrief.

Scoring & feedback

Evaluate the performance against the drill's Rubric signals and the Framework structure. Produce: an overall rating (e.g., Needs work / Solid / Strong), per-criterion notes (what was good, what was missing), and 2–3 concrete, actionable improvements.
Only after scoring, optionally show the Model answer (the ideal call flow or answer sketch) as a reference — never before.
Be specific and kind: quote what the learner actually said on the call or in the answer. The aim is improvement, not a grade.
Watch for the classic misses these rubrics encode:
- Cold call: no permission-based opener, pitching before earning interest, talking past the prospect's objection, no clear ask/CTA, sounding scripted instead of conversational, not handling the brush-off.
- Discovery: pitching instead of asking, closed questions only, not uncovering pain or its business impact, no quantifying ("how much is that costing you?"), failing to confirm next steps, talking more than listening.
- Behavioral: not using STAR, no real story, centering "we" instead of your actions, omitting the result/number, no resilience example (sales lives on rejection).
- Metrics & process: can't define the funnel or basic metrics (activity → pipeline → quota, conversion rates), no number sense, doesn't connect daily activity to results, treats a single metric uncritically.
- Motivation & company fit: generic answers ("I like talking to people"), no research on the company/product, no coachability or hunger, can't explain why sales or why here.

Repeat & improve loop

Each session draws a fresh drill (avoid immediate repeats) so practice — and the role-play scenarios — aren't memorized.
Track which rubric criteria the learner repeatedly misses across sessions and surface a "focus next on…" summary (e.g., "you often pitch before you've earned the right" or "you skip quantifying the pain in discovery").
Tie into the program (../program.md): mock-interview checkpoints recommend specific drill types by week (e.g., behavioral and motivation/fit early; cold-call and discovery role-plays and metrics/process before the job hunt, since SDR/AE interviews almost always include a live role-play).

Guardrails

The AI is a coach, not a gatekeeper — encourage retrying the call. A blown cold open is normal; the rep who reruns it is the one who improves.
Stay in character during role-plays; only break to debrief after the call ends. Don't rescue the learner by feeding lines mid-call.
Never expose the rubric/model answer before the learner attempts the drill.
Keep feedback grounded in the rubric; don't invent criteria.

Inputs the AI uses (already in the repo)

The drill prompt (the question, or the role-play scenario and prospect persona), its Framework (ideal call flow or answer structure), Model answer (reference call/answer), and Rubric (evaluation signals) — from the five drill files in this folder.
No additional data needed; the content is the spec.