🎙️ InterviewPrep — The Intelligent AI Mock Interview Platform

InterviewPrep is a next-generation, AI-powered mock interview platform designed to provide an ultra-realistic, adaptive interview experience. It goes beyond generic chatbots by parsing your resume, understanding your background, conducting a live audio-interactive interview, and adapting the difficulty of its questions in real-time based on your answers and confidence.

---

✨ Key Features

• 📄 Multimodal Resume Parsing: Upload a PDF, DOCX, or TXT resume. The system uses pdfplumber and spaCy NLP to extract your skills, experience, and infer your optimal job role.
• 🧠 Adaptive FSM Orchestration: Powered by a Finite State Machine (FSM), the interviewer dynamically adjusts topic and difficulty. If you struggle, it pivots to foundational questions; if you excel, it probes deeper into complex concepts.
• 🗣️ Real-Time Audio Interactivity: Speak your answers naturally. Utilizing Groq's whisper-large-v3-turbo for lightning-fast speech-to-text, paired with local DSP (librosa) to gauge speech rate and hesitation.
• ⚡ Dual-Eval Scoring Pipeline: To achieve sub-second conversational latency, answers are immediately evaluated using a synchronous keyword-matching heuristic to drive the conversation forward, while an asynchronous LLM evaluates technical depth in the background.
• 📊 Multi-Dimensional Coaching: Receive a comprehensive feedback report scoring you on Technical Correctness, Depth, Communication, Confidence, and Engagement, complete with actionable advice tied to your exact transcript.
• 🎯 Semantic Job Matching: Uses local sentence-transformers and a ChromaDB vector store to recommend real-world job roles that match your newly assessed skills and resume profile.

---

🛠️ Tech Stack

Frontend:

• Framework: Next.js 16 (App Router) + React 19
• State Management: Zustand (for lightweight, global interview state)
• Styling: TailwindCSS v4, shadcn/ui, Recharts

Backend & AI:

• API Framework: FastAPI + Uvicorn + Pydantic
• LLM Engine: Groq API (llama-3.3-70b-versatile for ultra-low latency reasoning)
• Speech & Audio: Groq Whisper (whisper-large-v3-turbo) + librosa
• Vector Database: ChromaDB (Local SQLite-backed)
• Embeddings: sentence-transformers (all-MiniLM-L6-v2)
• Data Store: In-memory session tracking with Redis as an optional cache layer
• Agents: LangChain + Custom FSM orchestrator

---

🏗️ Architecture & Data Flow

graph TD
    A[Candidate Resume] -->|Upload| B(Resume Parser & NLP)
    B -->|Extract Skills| C(Role Inference Agent)
    C -->|Initialize State| D{Interview Orchestrator FSM}
    
    D -->|Generate Q| E[Frontend UI]
    E -->|Audio/Text Answer| F(Fast-Path Eval)
    
    F -->|Transcribe| G[Whisper STT]
    F -->|Heuristics| H[Librosa DSP]
    
    G --> I[Keyword Matcher]
    I -->|Sync Result| D
    
    G -.->|Async Eval| J[Llama 3.3 Depth Eval]
    J -.->|Background Update| D
    
    D -->|Session End| K(Feedback Agent)
    K --> L[Final Report & Scores]
    
    B -.->|Profile Embedding| M[(ChromaDB Vector Store)]
    M -->|Semantic Match| N[Job Recommendations]
    N --> L

Loading

---

🚀 Quick Start (Local Development)

The platform is designed to run locally with minimal external dependencies. You do not need Docker to spin this up for the hackathon.

1. Prerequisites

• Node.js 20+
• Python 3.10+
• A Groq API Key (Get one free at console.groq.com)

2. Environment Setup

Clone the repository and create a .env file in the root:

git clone https://github.com/ASWD13/AI-mock-interview-tool.git
cd AI-mock-interview-tool
echo "GROQ_API_KEY=your_groq_api_key_here" > .env

3. Backend Setup

Install Python dependencies and download the required NLP models:

pip install -r requirements.txt
python -m spacy download en_core_web_sm

4. Seed the Vector Database

Populate the local ChromaDB with sample interview questions and job recommendations:

python scripts/seed_jobs.py
python scripts/seed_questions.py

5. Start the Services

Open two terminal windows.

Terminal 1 (Backend):

uvicorn backend.main:app --reload --port 8001

Terminal 2 (Frontend):

cd frontend
npm install
npm run dev

Navigate to http://localhost:3000 to start your mock interview!

---

💡Highlights & Talking Points

• Stateful Orchestration over Chatbots: Unlike a standard LLM wrapper that forgets context, InterviewPrep maintains a rigid Finite State Machine. It tracks rolling metrics (confidence, technical score, topics visited) to make deterministic decisions about when to escalate difficulty or gracefully shift topics.
• Latency-Obsessed Design: Live voice interviews feel unnatural with high latency. We separated evaluation into a "Fast-Path" (keyword extraction + audio heuristics) that unblocks the UI instantly, and a "Slow-Path" (deep LLM analysis) that updates the final score asynchronously.
• Multimodal Heuristics: We don't just judge what you say, but how you say it. By piping audio through librosa before transcription, we measure RMS energy (confidence) and zero-crossing rates (hesitation) to give nuanced behavioral feedback.

(Note: Advanced computer vision features using MediaPipe and DeepFace are implemented in the services/vision_analyzer.py module, but are currently disabled in the synchronous API flow to preserve sub-second response times for the demo.)

---

📄 License

This project is licensed under the MIT License.