🤖 AcePrep – AI Interview Practice Partner

Assignment Submission for Eightfold.ai > Problem Statement 2: Interview Practice Partner

AcePrep is a fully agentic, voice-first mock interview platform. It moves beyond static text generation by offering a real-time, conversational voice agent that adapts its tone to the user’s chosen interview style and maintains a structured, natural interview flow, generating detailed feedback upon completion.

---

🔗 Quick Links

• Live Demo: https://ace-prep-umber.vercel.app/

---

🧠 Design Decisions & Architecture

1. Why VAPI.ai for Voice Orchestration?

• Reasoning: Building a raw WebSocket layer for voice introduces significant latency and state management complexity.
• Decision: I chose VAPI as the orchestration layer to minimize "Turn-Taking" latency. This ensures the AI doesn't interrupt the user while thinking, which is critical for the "Conversational Quality" criteria.

2. Gemini 2.0 Flash for Intelligence

• Reasoning: Voice agents require sub-second inference speeds.
• Decision: Gemini 2.0 Flash was selected over GPT-4o for its superior speed-to-cost ratio and its ability to adhere strictly to JSON schemas for the interview generation phase.

3. Serverless Next.js + Firebase

• Reasoning: The assignment requires a public demo.
• Decision: A serverless architecture ensures the application scales automatically during the demo phase without server maintenance. Firebase Firestore provides real-time data syncing, allowing the dashboard to update immediately once an interview is generated.

---

🤖 Agentic Behavior

The agent includes the following actual behaviours as required by the assignment:

• Tone adaptation → fast, friendly, or detailed
• Structured question progression (increasing difficulty)
• Tech-stack-aware question generation
• Role & level contextualization
• JSON-controlled generation with no hallucinations
• Backend structured AI feedback after interview

---

🚀 Features

✅ Dynamic Interview Generation

The voice agent acts as a recruiter first. It collects:

• Role (e.g., Data Analyst, Frontend Dev)
• Tech Stack (e.g., React, SQL, Python)
• Experience Level
• Interview Style (Fast vs. Friendly)

✅ Real-Time Voice Interview

• Latency Optimization: optimized for natural pauses.
• Context Awareness: The agent remembers previous answers (context window) to ask relevant follow-up questions.

✅ Automated Feedback Loop

Once the call ends, a background process triggers: 1.  Transcript analysis. 2.  Scoring based on Technical Accuracy, Communication, and Problem Solving. 3.  Generation of a "Strengths & Weaknesses" report.

---

🖼️ Screenshots

Sign In Page

Sign Up Page

Home Page

Interview Page

Feedback Page

---

🏗 Project Structure

.
├── app/
│   ├── api/
│   │   ├── generate/
│   │   │   └── route.tsx          # AI generation logic (Gemini)
│   │   └── feedback/
│   │       └── route.tsx          # Feedback generation logic
│   ├── (root)/
│   │   └── dashboard/
│   │       └── page.tsx           # Dashboard (Your vs Public interviews)
│   ├── interview/
│   │   └── [id]/
│   │       └── page.tsx           # Interview Room (Agent & Vapi)
│   ├── globals.css                # Global styles
│   ├── layout.tsx                 # Root layout
│   └── page.tsx                   # Landing page
├── components/
│   ├── ui/                        # Shadcn UI components (Button, Card, etc.)
│   ├── Agent.tsx                  # Vapi Voice Client
│   ├── DisplayTechIcons.tsx       # Tech stack icon renderer
│   └── InterviewCard.tsx          # Dashboard card component
├── constants/
│   └── index.ts                   # Static data & Prompts
├── firebase/
│   └── admin.ts                   # Firebase Admin configuration
├── lib/
│   ├── actions/
│   │   ├── auth.action.ts         # Authentication logic
│   │   └── general.action.ts      # Firestore fetch/save logic
│   ├── utils.ts                   # Helpers (Icon mapping, cn)
│   └── vapi.sdk.ts                # Vapi SDK init
└── types/
    └── index.d.ts                 # TypeScript interfaces (Interview, User)

🛠️ Tech Stack

• Frontend: Next.js 14 (App Router), TailwindCSS, ShadCN UI
• AI Logic: Google Gemini 2.0 Flash
• Voice Pipeline: VAPI (Voice Agent Protocol Interface)
• Database: Firebase Firestore
• Auth: Custom Firebase Authentication (Session Cookies)
• Deployment: Vercel

---

🔌 Setup Instructions

1. Clone the repository

git clone https://github.com/ananyaarramalla/AcePrep-Agentic-Assignment.git
cd AcePrep-Agentic-Assignment

2. Install Dependencies

npm install

3. Environment Variables

Create a .env.local file in the root directory and add the following keys:

NEXT_PUBLIC_VAPI_PUBLIC_KEY=your_vapi_key
NEXT_PUBLIC_VAPI_WORKFLOW_ID=your_workflow_id
FIREBASE_PROJECT_ID=your_firebase_id
FIREBASE_CLIENT_EMAIL=your_email
FIREBASE_PRIVATE_KEY=your_private_key
GOOGLE_API_KEY=your_gemini_key

4. Run Locally

npm run dev

Visit http://localhost:3000 to view the application.

🧪 Testing The Flow

• 1.  Sign In: Access the dashboard via the login page.
• 2.  Generate: Click "Start New Interview". Speak to the voice agent to set your preferences (Role, Tech Stack, etc.).
• 3.  Verify: Check the Firestore database to see the JSON object created with your specific questions.
• 4.  Interview: Enter the interview room and complete the mock session with the AI agent.
• 5.  Feedback: Wait ~10 seconds after ending the call to see the generated feedback card with scores and improvements.

---

Author: Ananya Arramalla

Submitted for Eightfold.ai AI Agent Assignment