interview-prep

name	interview-prep
description	Prepare for job interviews with tailored questions, STAR answers, and company research. Builds interactive web apps for practice.

Job Interview Prep Kit

Prepare for interviews with company-specific research, behavioral story banks, technical frameworks, and salary negotiation scripts. Always delivers as an interactive web app.

When to Use

• Upcoming interview, mock question practice, salary negotiation prep, behavioral story coaching

When NOT to Use

• Resume creation (resume-maker), broad career research (deep-research)

Step 0: Ask the User — Mode Selection

Before doing anything else, ask the user:

1. What role and company are you interviewing for?
2. Which prep mode do you want?
   • Voice mock interview — AI-powered voice conversation simulating a real interview (uses OpenAI Realtime API for speech-to-speech)
   • Text mock interview — Chat-based interview simulator where the AI asks questions and gives feedback in real time
   • Interview prep dashboard — A reference website with expected questions, company intel, story frameworks, and negotiation scripts

All three modes are built as web apps deployed as artifacts. Do not output raw markdown — always build an interactive app.

Step 1: Research the Role and Company

Before building anything, use webSearch to gather real intel. This data populates whichever mode the user chose.

Source	Query	What you get
Glassdoor	site:glassdoor.com [Company] interview questions [role]	Actual questions asked, by round. Filter to last 6 months.
Blind	site:teamblind.com [Company] interview OR [Company] onsite	Unfiltered loop structure, which rounds matter, bar-raiser tells, comp bands
levels.fyi	site:levels.fyi [Company]	Real comp by level + location. Critical for negotiation anchoring.
LeetCode Discuss	site:leetcode.com/discuss [Company] [role]	Tagged coding problems actually asked
LinkedIn	[Company] [team name] → recent posts from hiring manager	What they're shipping, what they celebrate, vocabulary they use
Eng blog / Newsroom	[Company] engineering blog	System design context. Mentioning their published architecture in an interview is a strong signal.

For Amazon specifically: each interviewer is assigned 1-3 Leadership Principles to probe. Map stories to LPs before the loop.

Step 2: Build the Web App

Mode A: Voice Mock Interview

Build a React web app that uses the OpenAI Realtime API for voice-to-voice interview simulation. The app should:

• Connect to OpenAI's Realtime API via WebSocket for low-latency speech-to-speech
• Configure the AI persona as an interviewer for the specific role/company
• Seed the system prompt with the researched questions from Step 1
• Include a question queue panel showing upcoming questions (behavioral, technical, situational) drawn from Glassdoor/Blind research
• Show a live transcript panel so the user can review what they said
• After each answer, provide written feedback on: STAR structure, specificity, quantification, and areas to improve
• Include a timer per question (recommended 2-3 min per behavioral, 15-20 min per system design)
• End with a scorecard summarizing performance across categories

// OpenAI Realtime API connection pattern
const pc = new RTCPeerConnection();
const audioEl = document.createElement("audio");
audioEl.autoplay = true;

// Add local audio track for user's microphone
const ms = await navigator.mediaDevices.getUserMedia({ audio: true });
pc.addTrack(ms.getTracks()[0]);
pc.ontrack = (e) => { audioEl.srcObject = e.streams[0]; };

// Connect to OpenAI Realtime
const tokenRes = await fetch("/api/realtime-session", { method: "POST" });
const { client_secret } = await tokenRes.json();
const offer = await pc.createOffer();
await pc.setLocalDescription(offer);

const sdpRes = await fetch("https://api.openai.com/v1/realtime?model=gpt-4o-realtime-preview", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${client_secret.value}`,
    "Content-Type": "application/sdp",
  },
  body: offer.sdp,
});
await pc.setRemoteDescription({ type: "answer", sdp: await sdpRes.text() });

The backend route (/api/realtime-session) creates an ephemeral token via:

const r = await fetch("https://api.openai.com/v1/realtime/sessions", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.OPENAI_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    model: "gpt-4o-realtime-preview",
    voice: "verse",
    instructions: `You are an interviewer for [Role] at [Company]. Ask behavioral and technical questions one at a time. After the candidate answers, give brief feedback, then move to the next question. Questions: [insert researched questions from Step 1]`,
  }),
});
const data = await r.json();
// Return data.client_secret to the frontend

Mode B: Text Mock Interview

Build a React chat app that simulates a text-based interview. The app should:

• Present a chat interface styled like a messaging app
• AI asks one question at a time from the researched question bank
• After each user response, AI provides inline feedback (STAR structure check, specificity score, improvement suggestions)
• Include a sidebar with: question categories (behavioral / technical / situational), progress tracker, tips for the current question type
• Use the OpenAI Chat Completions API (not Realtime) via a backend route
• End with a summary report scoring each answer

System prompt for the interviewer AI:

You are interviewing a candidate for [Role] at [Company]. Ask questions one at a time from this list: [researched questions]. After each answer, give brief, constructive feedback focusing on: specificity, use of "I" vs "we", quantified results, and STAR structure. Then ask the next question. Be professional but conversational.

Mode C: Interview Prep Dashboard

Build a polished single-page React app with these sections:

1. Company Intel Panel — loop structure, recent news (from web search), team info, culture notes
2. Question Bank — categorized tabs (Behavioral, Technical, System Design, Culture Fit) with actual questions from Glassdoor/Blind research. Each question expandable to show:
   • Why they ask this
   • Framework to use (STAR, CAR, SOAR, STAR+L)
   • Example strong answer structure
3. Story Builder — interactive 6-story matrix (Leadership, Conflict, Failure, Ambiguity, Impact, Scope Creep). User fills in their STAR components; the app validates completeness and flags missing quantification
4. Comp & Negotiation — levels.fyi data, target/walk-away ranges, scripted negotiation lines
5. Questions to Ask — curated list with context on what each reveals
6. Countdown Timer — if the user has an interview date, show days remaining with a suggested daily prep schedule

Behavioral Story Frameworks

STAR is the baseline. Know the variants:

• STAR — Situation (1-2 sent) → Task (1 sent) → Action (60% of airtime, use "I" not "we") → Result (quantified)
• CAR — Challenge → Action → Result. Tighter; better for rapid-fire rounds.
• SOAR — Situation → Obstacle → Action → Result. Use when the story's value is in what you overcame.
• STAR+L — Append Learning. Mandatory for failure questions.

Build a 6-story matrix, not a script per question. One well-told story covers 3-4 question variants.

Category	Prepare 1 story each	Maps to questions like
Leadership	Stepped up without authority	"Influenced without authority," "led a project"
Conflict	Disagreed with manager/peer, resolved	"Difficult coworker," "disagree and commit"
Failure	Owned a mistake, quantify damage + fix	"Project that failed," "missed a deadline"
Ambiguity	Decided with incomplete info	"Moved fast," "prioritized under uncertainty"
Impact	Your single biggest measurable win	"Most proud of," "biggest accomplishment"
Scope creep	Did more than asked	"Exceeded expectations," "ownership"

"Tell me about yourself": Present (current role + one win) → Past (how you got here, 1 pivot) → Future (why this role is the logical next step). 90 seconds.

Technical Frameworks

System design — RADIO:

• Requirements — Functional + non-functional. Ask: scale? latency target? (~15% of time)
• Architecture — Boxes and arrows. Client, API layer, services, data stores, queues. (~20%)
• Data model — Entities, fields, relationships. (~15%)
• Interface — API contracts. REST vs GraphQL vs gRPC. (~15%)
• Optimizations — Caching, CDN, sharding, read replicas. Senior signal lives here. (~35%)

Coding pattern recognition: ~75% of LeetCode-style questions reduce to: sliding window, two pointers, BFS/DFS, binary search on answer, heap for top-K, DP (1D/2D), union-find, monotonic stack.

Salary Negotiation

1. Never give a number first. "I'd want to learn more about the role before discussing comp — what range did you have budgeted?"
2. Only negotiate after a yes. Once they want to hire you.
3. BATNA is everything. Competing offer > current job > nothing.
4. Anchor with data. levels.fyi for exact company + level + location. Quote 75th percentile.
5. Negotiate the package, not the base. Signing bonus, equity refresh, start date, remote days, title.
6. The reframe script: "I'm really excited about this. Based on levels.fyi data for [level] at [company], I was expecting something closer to [X]. Is there flexibility?"
7. Silence is a tool. State your number. Stop talking.
8. Get it in writing.

Questions to Ask Interviewers

• "What's the one thing that, if it goes wrong in the next 6 months, keeps you up at night?"
• "What did the last person who was great in this role do differently?"
• "What's something the team tried recently that didn't work?"
• "How are decisions made when engineering and product disagree?"
• To the hiring manager: "What would make you say 'I'm so glad we hired them' at the 6-month mark?"

Best Practices

1. 6 stories, not 30 answers — depth beats breadth
2. Quantify or it didn't happen — "cut p99 from 340ms to 45ms" > "improved performance"
3. Practice out loud — written answers don't survive contact with your mouth
4. "I don't have a great example" is fine — better than a weak story. Pivot: "The closest I have is..."
5. Failure questions: own it fully — hedging is the most common disqualifier

Limitations

• Voice mode requires user to provide their OpenAI API key (for Realtime API access)
• Comp data lags reality by ~3-6 months; cross-reference Blind for recency
• Company-specific intel quality depends on how much employees post publicly
• Cannot access internal question banks or recruiter portals