Interview Prep App

A modern, full-stack interview preparation app powered by multiple LLM providers (OpenAI, Anthropic, Gemini, Ollama) and an in-memory vector database for semantic job description search (RAG).

---

🚀 Features

• Multi-provider LLM support: OpenAI, Anthropic (Claude), Gemini, Ollama (local models)
• Configurable LLM settings: Temperature, max tokens, jailbreak protection, etc.
• Job description RAG: Semantic search and enrichment for job descriptions
• Interview setup wizard: Step-by-step configuration for your interview session
• Prompting techniques: Zero-shot, few-shot, chain-of-thought, role-based, and more
• Modern UI: Built with React, MUI, and Next.js
• In-memory vector DB: No external DB required for job corpus

---

🗺️ Application Flow (Mermaid Diagram)

flowchart TD
    A[User opens app] --> B[Setup Wizard]
    B --> C[Select LLM Provider & Model]
    C --> D[Configure LLM Settings]
    D --> E[Select Interview Topic]
    E --> F[Choose Knowledge/Background]
    F --> G[Set Difficulty, Persona, Response Length]
    G --> H[Enter or Generate Job Description]
    H --> I[Select Prompting Technique]
    I --> J[Start Interview]
    J --> K[User submits prompt/question]
    K --> L{RAG: Search Job Corpus?}
    L -- Match Found --> M[Retrieve Job Description]
    L -- No Good Match --> N[Generate with LLM]
    M & N --> O[Prepare System Prompt]
    O --> P[Send to Selected LLM Provider]
    P --> Q[LLM Generates Response]
    Q --> R[Show Response in Chat UI]
    R --> S{Continue?}
    S -- Yes --> K
    S -- No --> T[End Session]

    subgraph Provider Selection
      C
      D
    end
    subgraph Interview Setup
      E
      F
      G
      H
      I
    end
    subgraph RAG
      L
      M
      N
    end
    subgraph LLM Interaction
      O
      P
      Q
    end

Loading

---

🛠️ Getting Started

1. Clone the repository

git clone <your-repo-url>
cd interview-prep-app

2. Install dependencies

yarn install

3. Set up environment variables

Copy .env.example to .env and fill in your API keys:

cp .env.example .env

Required variables:

• OPENAI_API_KEY (for OpenAI models)
• ANTHROPIC_API_KEY (for Anthropic/Claude models)
• GEMINI_API_KEY (for Gemini models)
• OLLAMA_BASE_URL (optional, for custom Ollama server URL; defaults to http://localhost:11434)

Ollama does not require an API key but must be running locally or accessible at the specified URL.

4. Run the app in development

yarn dev

Visit http://localhost:3000 in your browser.

---

⚙️ LLM Provider Support

• OpenAI: GPT-4o, GPT-4, GPT-3.5, etc. (requires API key)
• Anthropic: Claude 3, Claude 3.5, etc. (requires API key)
• Gemini: Gemini 1.5, Gemini Pro, etc. (requires API key)
• Ollama: Local models (e.g., Mistral, Llama2, etc.)
  • Ollama must be running locally (ollama serve)
  • Models must be pulled (e.g., ollama pull mistral)

---

🧠 Vector Database (RAG)

• Uses imvectordb for in-memory semantic search
• Corpus is stored in job_corpus.json (auto-saved/loaded)
• No external DB or cloud service required

---

🧑‍💻 Development

Linting & Type Checking

yarn lint        # Run ESLint
yarn type-check  # Run TypeScript compiler

Formatting

yarn format      # Run Prettier

Build for Production

yarn build

Start in Production

yarn start

---

🐛 Troubleshooting

• Ollama returns empty messages:
  • Ensure Ollama is running and the model is pulled
  • Try a simple prompt via the Ollama CLI to verify
  • Check logs for errors in src/lib/llm-ollama.ts
• API key errors:
  • Make sure your .env file is set up and keys are valid
• Vector DB issues:
  • job_corpus.json must be writable by the app

---

🤝 Contributing

1. Fork the repo and create your branch
2. Make your changes (lint and type-check before PR!)
3. Submit a pull request

---

📄 License

MIT

---

🛠️ Configuration (src/config/config.ts)

The app is highly configurable via src/config/config.ts. You can customize:

• Models:
  • Add or remove LLM models/providers (OpenAI, Anthropic, Gemini, Ollama, etc.)
  • Example:

    models: [
      { id: 'gpt-4o', label: 'GPT-4o', provider: 'openai' },
      { id: 'mistral', label: 'Mistral (Ollama)', provider: 'ollama' },
      // ...
    ]

• Prompt Options:

  • Define different prompting techniques for the interview session. Each option has an id, label, and description.

  • Available options:

    ID	Label	Description (Purpose)
    zero-shot	Zero-Shot Prompting	No examples, tests adaptability and core skills.
    few-shot	Few-Shot Prompting	Provides example Q&A pairs to guide expectations.
    chain-of-thought	Chain-of-Thought	Encourages step-by-step reasoning and problem-solving.
    role-based	Role-Based Prompting	Switches interviewer roles to assess breadth and adaptability.
    instruction-based	Instruction-Based Prompt	Highly structured, standardized interviews for consistent evaluation.

  • Example config for a prompt option:

    {
      id: 'zero-shot',
      label: 'Zero-Shot Prompting',
      description: `Context: You are an expert technical interviewer in an LLM interview arena.\nObjective: Conduct a technical interview without any examples or prior context.\nStyle: Professional and adaptable.\nTone: Analytical yet encouraging.\nAudience: Interview candidates of varying experience levels.\nResponse: Ask relevant technical questions, evaluate responses in real-time, and provide constructive feedback. Adapt your questions based on the candidate's responses and skill level demonstrated. Focus on core competencies and problem-solving abilities.`
    }

  • How to add your own:

    • Add a new object to the promptOptions array in src/config/config.ts with a unique id, a descriptive label, and a detailed description.
    • The description can include context, objectives, style, tone, audience, and response guidelines for the interviewer LLM.

  • Detailed breakdown:

    • Zero-Shot Prompting:
      • Purpose: Conduct interviews without providing examples or prior context. Focuses on adaptability and core skills.
      • Use case: Simulate a real interview where the candidate must respond without hints.
      • Example config: See above.
    • Few-Shot Prompting:
      • Purpose: Guide the interview with example Q&A pairs, helping the model understand the expected format and depth.
      • Use case: Provide a few sample questions/answers to set expectations for the candidate.
      • Example config:

        {
          id: 'few-shot',
          label: 'Few-Shot Prompting',
          description: `Context: You are an expert technical interviewer with access to example Q&A pairs.\nObjective: Use example patterns to guide the interview process.\nStyle: Structured and consistent.\nTone: Educational and supportive.\nAudience: Interview candidates seeking clear expectations.\nResponse: Here are three example Q&A pairs:\n1. Q: 'Explain dependency injection?' A: [Clear, concise answer]\n2. Q: 'What is SOLID?' A: [Well-structured response]\n3. Q: 'Describe microservices' A: [Comprehensive explanation]\nUse these patterns to formulate similar questions and evaluate responses accordingly.`
        }

    • Chain-of-Thought:
      • Purpose: Encourage step-by-step reasoning and problem-solving by prompting the candidate to explain their thought process.
      • Use case: Technical or architectural challenges where reasoning is as important as the answer.
      • Example config:

        {
          id: 'chain-of-thought',
          label: 'Chain-of-Thought',
          description: `Context: You are an expert interviewer focused on understanding candidate reasoning.\nObjective: Guide candidates through complex problem-solving scenarios.\nStyle: Systematic and analytical.\nTone: Collaborative and inquisitive.\nAudience: Candidates solving technical or architectural challenges.\nResponse: Present a problem, then guide with prompts like:\n1. 'What's your initial approach?'\n2. 'What assumptions are you making?'\n3. 'What trade-offs do you see?'\n4. 'How would you test this solution?'\nEncourage detailed explanations at each step.`
        }

    • Role-Based Prompting:
      • Purpose: Assess candidates from different professional perspectives (e.g., technical lead, product manager, HR).
      • Use case: Switch between roles to evaluate a candidate's breadth and adaptability.
      • Example config:

        {
          id: 'role-based',
          label: 'Role-Based Prompting',
          description: `Context: You can embody different interviewer personas.\nObjective: Assess candidates from various professional perspectives.\nStyle: Adaptive and role-specific.\nTone: Varies by role (e.g., technical lead: detailed, HR: interpersonal).\nAudience: Candidates being evaluated across different dimensions.\nResponse: Switch between roles such as:\n1. Technical Lead: Deep technical questions\n2. Product Manager: System design and trade-offs\n3. Team Lead: Leadership and collaboration\n4. Architect: High-level design principles\nAdjust questioning style and focus based on the current role.`
        }

    • Instruction-Based Prompting:
      • Purpose: Conduct highly structured interviews with specific instructions and evaluation criteria.
      • Use case: Screen candidates with standardized questions and formats for consistent evaluation.
      • Example config:

        {
          id: 'instruction-based',
          label: 'Instruction-Based Prompting',
          description: `Context: You are a structured interviewer following specific guidelines.\nObjective: Conduct highly focused, consistent interviews.\nStyle: Clear and methodical.\nTone: Direct and constructive.\nAudience: Candidates requiring specific evaluation criteria.\nResponse: Follow these instructions:\n1. Start with role-specific screening questions\n2. Present coding/design challenges with clear requirements\n3. Request specific formats for answers (e.g., 'Explain in pseudocode')\n4. Evaluate against predefined criteria\n5. Provide structured feedback\nMaintain consistency while allowing for candidate-specific adaptations.`
        }

• Topics:
  • Add interview topics (e.g., "Web Applications", "Prompt Engineering", "Venture Capital")
  • Each topic can have:
    • knowledgeOptions: User backgrounds/levels for the topic
    • interviewerGuidelines: Technical and behavioral criteria for evaluation

To extend:

• Add new models to the models array
• Add new topics or knowledge levels to the topics array
• Add or edit prompt options for new interview styles
• Customize interviewer guidelines for your domain

Changes to config.ts are picked up automatically on restart.