interview.me

AI-powered voice interview practice platform. Paste a job posting URL, get a realistic mock interview tailored to the role, and receive a detailed performance report with scored feedback.

Built for the DigitalOcean Gradient AI Hackathon.

How It Works

%%{init: {'flowchart': {'htmlLabels': true, 'fontSize': 13}}}%%
flowchart LR
    A["Paste job URL"] --> B["Enrich context"]
    B --> C["Voice interview"]
    C --> D["Performance report"]

    B -.- B1["Scrape posting via Jina Reader
    Extract requirements, tech stack
    Search for real interview questions"]

    C -.- C1["AI interviewer asks 4-5 questions
    ElevenLabs streaming TTS + browser STT
    Responses streamed via SSE"]

    D -.- D1["LLM evaluates full transcript
    Scored feedback across 5 categories
    Question-by-question breakdown"]

Loading

Under the hood

• Context enrichment — Jina Reader scrapes the job posting while Serper searches for past interview questions and company culture. An LLM extracts structured data (requirements, tech stack, seniority) and generates a role-specific interviewer persona. Deep enrichment continues in the background, pulling interview questions from Glassdoor/Blind snippets.
• Interview runtime — Each chat turn gets injected system messages: turn guidance (behavioral → technical → closing), candidate behavior detection (blocks hijacking or early farewells), and a hard brevity constraint (≤40 words) in the recency position. Per-turn token limits prevent mid-sentence truncation.
• Voice loop — Frontend state machine (AI_SPEAKING → LISTENING → PROCESSING → AI_SPEAKING). TTS uses ElevenLabs with streaming playback via MediaSource API — audio starts playing as chunks arrive rather than waiting for the full MP3. Falls back to browser Web Speech API when ElevenLabs is unavailable. STT uses browser Web Speech API.
• Report — Streamed on-demand via SSE. The LLM evaluates the transcript and outputs scored feedback with a question-by-question breakdown.

Project Structure

interview.me/
├── frontend/       React + Vite + Tailwind CSS (voice UI)
├── backend/        Express + PostgreSQL + DO Gradient AI
├── .do/            App Platform deployment spec
└── package.json    npm workspaces root

Getting Started

Quick Start (Docker)

The fastest way to run the full stack — no Node.js or PostgreSQL setup needed.

1. Install Docker Desktop
2. Clone the repo and create a .env file in the project root:

DO_MODEL_ACCESS_KEY=your-key-from-digitalocean
SERPER_API_KEY=your-serper-key
ELEVENLABS_API_KEY=your-elevenlabs-key
ELEVENLABS_VOICE_ID=your-voice-id

3. Run:

docker compose up --build

4. Open http://localhost:8080

Manual Setup

Prerequisites

• Node.js 20+
• PostgreSQL (any recent version — 16, 17, or 18)
• DigitalOcean Model Access Key

1. Install PostgreSQL

macOS (Homebrew):

brew install postgresql
brew services start postgresql

Ubuntu/Debian:

sudo apt install postgresql
sudo systemctl start postgresql

2. Create the database

createdb interviewme

3. Install dependencies

npm install

4. Configure environment

cp backend/.env.example .env

Edit .env:

DATABASE_URL=postgresql://your-username@localhost:5432/interviewme
DO_MODEL_ACCESS_KEY=your-key-from-digitalocean
PORT=3001

# Optional: Context enrichment APIs
SERPER_API_KEY=your-serper-key

# Optional: ElevenLabs TTS (falls back to browser speech synthesis if not set)
ELEVENLABS_API_KEY=your-elevenlabs-key
ELEVENLABS_VOICE_ID=your-voice-id

To get a Model Access Key, go to DigitalOcean GenAI Inference and create a key.

5. Run

npm run dev

Frontend runs on http://localhost:5173, backend on http://localhost:3001. The Vite dev server proxies /api requests to the backend automatically. Database tables are created automatically on first startup.

Scripts

Command	Description
npm run dev	Start frontend + backend concurrently
npm run dev:frontend	Start frontend dev server only
npm run dev:backend	Start backend dev server only
npm run build:frontend	Production build frontend
npm run build:backend	Production build backend

Tech Stack

Frontend: React 19, Vite 8, TypeScript, Tailwind CSS v4, shadcn/ui, Motion, Web Speech API (STT), ElevenLabs streaming TTS

Backend: Express 5, PostgreSQL, OpenAI SDK (DO Gradient compatible), Server-Sent Events

AI: DigitalOcean Gradient AI — minimax-m2.5 for interview conversation, prompt generation, and report generation; openai-gpt-oss-120b for brief generation and search query planning

Enrichment: Jina Reader API (job page scraping), Serper API (interview question search)

Deployment: DigitalOcean App Platform (static site + service + managed database)

API Endpoints

Method	Path	Description
POST	/api/sessions	Create interview session with enrichment
GET	/api/sessions/:id	Get session + message history
POST	/api/chat	Send message, stream AI response (SSE)
POST	/api/sessions/:id/end	End session
GET	/api/sessions/:id/report/stream	Stream performance report (SSE)
GET	/api/sessions/:id/report	Get report (200 ready, 202 generating)
POST	/api/tts	Text-to-speech via ElevenLabs (streamed)