Interview Practice App

A single-page Interview Practice App for Software & AI Engineering interview preparation, built with Streamlit, OpenAI, and Pydantic.

Project Structure

.
├── app.py                      # Main Streamlit application entry point
├── requirements.txt            # Project dependencies
├── src/                        # Application source code
│   ├── config.py               # Settings and .env loading logic
│   ├── guards.py               # Input validation & per-session rate limiting
│   ├── interview_state.py      # Dataclasses for tracking turns and session status
│   ├── json_utils.py           # Robust JSON extraction and parsing from LLM outputs
│   ├── llm_client.py           # OpenAI client wrapper with retry logic and timeouts
│   ├── logging_setup.py        # Centralized loguru configuration
│   ├── pricing.py              # Token counting and cost estimation logic
│   ├── prompts.py              # Jinja2 template rendering and strategy mapping
│   ├── schemas.py              # Pydantic models for structured AI responses
│   └── ux_utils.py             # Helper functions for state management and error mapping
├── templates/                  # Jinja2 prompt templates
│   ├── system/                 # Strategy-specific system prompts
│   └── user/                   # User prompts for extraction, planning, and feedback
└── ...

Technical Architecture

• Streamlit Frontend: A modern, interactive web interface with sidebar controls for deep LLM parameter tuning.
• Jinja2 Templating Engine: All prompts (system and user) are managed as .j2 templates, allowing for modular, clean, and dynamic prompt construction.
• Pydantic Validation: Strict schema enforcement for JSON outputs (Interview Plans, Focus Areas, Final Feedback) to ensure reliability.
• Robust Error Mapping: A custom UX layer that translates backend exceptions (API timeouts, JSON failures, missing keys) into user-friendly guidance.
• State Management: Sophisticated session state handling to prevent double-click issues and maintain consistent interview flows.

Setup

1) Install dependencies

Run the following command in your terminal:

pip install -r requirements.txt

2) Configure environment variables

Create a local .env file from the example:

cp .env.example .env

Edit .env and provide your OpenAI API key:

OPENAI_API_KEY=your_openai_api_key_here

(Optional) Customize OPENAI_PRICING_INPUT_PER_1M and OPENAI_PRICING_OUTPUT_PER_1M for more accurate cost estimation.

3) Run the app

streamlit run app.py

Features

• Guided Mock Interview Flow: Structured 5-question interview experience (Start → Next → End) with full transcript management.
• Job Description Extraction: Automatically extract role titles and target focus areas from any pasted job description using AI.
• Advanced Prompt Engineering: Access to 10+ prompt strategies (Zero-shot, Few-shot, Self-refinement, Maieutic, etc.) selectable in real-time.
• UX Safety & Polish:
  • Reset Interview: Dedicated action to cleanly wipe session state and start fresh.
  • Settings Warning: Real-time notification if critical interview parameters are changed mid-session.
  • Single-Click UX: Optimized interaction flow for seamless question transitions.
• Tunable Generation Settings: Granular control over Model, Temperature, Top-p, Penalties, Timeout, and Retries.
• Approximate Cost Tracking: Real-time estimation of input/output tokens and USD cost based on configurable pricing models.
• Security Guardrails: Built-in input validation, basic prompt-injection heuristics, and session-based rate limiting (10 calls/min).

How to use

1. Preparation

• Fill in the Role title and Focus areas.
• Optional: Paste a Job description and click Extract role title and focus areas to let the AI populate the fields for you.

2. Guided Interview

• Configure your Model, Difficulty, and Persona in the sidebar.
• Click Start interview.
• Provide your answer in the text area and click Next question.
• After 5 turns (or earlier), click End interview & get feedback.
• View your Final feedback (text) and Final feedback (JSON).

3. Structured Tools

• Generate interview plan (JSON): Create a downloadable roadmap for your preparation.
• Download buttons: Export any generated JSON (Plans or Feedback) directly to your local machine.

Prompt strategies included

The app demonstrates various techniques from the "Prompt Engineering" field:

• Zero-shot: Direct instructions without examples.
• Few-shot: Providing style examples within the prompt.
• Chain-of-Thought / Maieutic: Forcing logical steps before reaching a conclusion.
• Self-refinement: An internal loop where the model critiques its own first draft.
• Least-to-most: Breaking down complex tasks into sequential sub-tasks.
• Delimiters: Using distinct markers to separate instructions from data.
• Condition-checking: Explicitly verifying requirements before proceeding.
• Generated knowledge: Prompting the model to recall relevant concepts first.
• JSON-only enforcement: Forcing structured outputs for tool integration.
• App critic: A self-assessment strategy where the AI reviews its own performance.

Security & Reliability

• Data Safety: Never paste secrets (API keys, passwords) into the chat. The app uses environment variables for the main API key.
• Stability: The app uses a retry mechanism and specific timeouts to handle API instability gracefully.
• Rate Limiting: A simple indicator in the UI tracks your usage to help stay within API quotas.

Disclaimer

• Token and cost estimates are approximate and intended for awareness only.
• Model behavior may vary depending on the chosen prompt strategy and parameters.