Interview Agent

An AI-powered mock interview coach. Practice technical and behavioral interviews with real-time feedback, resume parsing, and session tracking.

Live Demo: https://www.openagentsbox.com/interview

---

Features

• Mock Interviews — Conduct full mock interviews in text or voice mode. Choose interview type (technical, behavioral, system design) and difficulty.
• Resume Parsing — Upload a PDF resume and have it automatically parsed into structured data to personalize interview questions.
• Real-time Scoring — Get scored on each answer with detailed feedback after the session.
• Session History — Review past interview sessions, transcripts, and performance reports.
• Job Match Analysis — Paste a job description to get a match analysis against your resume.
• Multi-model Support — Works with Anthropic (Claude) and OpenAI (GPT) models. Switch models per role in settings.

---

Quick Start (Online)

Visit https://www.openagentsbox.com/interview — no installation needed.

1. Click the key icon (top-right) to set your Anthropic or OpenAI API key.
2. Add a company and create a new interview session.
3. Start practicing.

Your API key is stored locally in your browser and never sent to our servers except to forward to the LLM provider.

---

Self-hosting (Docker)

Run the full stack locally with a single Docker image (frontend + backend bundled together).

Requirements: Docker

1. Build

docker build -f Dockerfile.standalone -t interview-standalone .

2. Set up environment variables

Copy the example env file and fill in your API keys:

cp .env.example .env

Edit .env:

STORAGE_MODE=local

API keys are not required in .env. Users enter their own Anthropic or OpenAI key directly in the app UI — keys are stored in the browser and never saved on the server.

3. Run

Windows (PowerShell):

docker run -p 8000:8000 --env-file .env -v "${PWD}\data:/app/data" interview-standalone

macOS / Linux:

docker run -p 8000:8000 --env-file .env -v "$(pwd)/data:/app/data" interview-standalone

Open http://localhost:8000/interview in your browser.

Data persistence: All your data (resume, companies, sessions, reports) is stored in the data/ folder on your local machine. Deleting or recreating the container will not affect your data as long as you keep the data/ folder and mount it with -v.

---

Agent Core Concept

The interview agent is built around a tool-loop pattern inspired by Claude Code's task tracking:

The LLM is given a pre-generated task list (interview questions/topics) at session start. On every user reply, the agent runs a while True loop:

1. check_task — reads the current task list and which task is in_progress
2. The LLM conducts the interview, asking follow-up questions as needed
3. update_task — when the candidate has sufficiently answered, mark the task complete with a score (1–10) and a private evaluation note
4. Move to the next task — repeat until all tasks are completed
5. Return final reply with all_finished = True → frontend shows the session end screen

This gives the agent a persistent, inspectable sense of progress across multi-turn conversations without relying on the LLM's memory alone.

---

License

CC BY-NC 4.0 — Free to use and modify for non-commercial purposes. Attribution required.

---

Tech Stack

• Frontend: React + TypeScript + Vite + Tailwind CSS
• Backend: FastAPI + LiteLLM + Instructor
• LLM: Anthropic Claude / OpenAI GPT (user-provided API key)