Gen AI Interview Prep

A free, open-source, self-hosted tool for Gen AI developer interview preparation.

Browse 690+ realistic interview questions across 15 Gen AI domains, each paired with a curated ideal answer — the kind of answer that would impress a senior interviewer. No timers, no scores, no pressure.

Bring Your Own Key (BYOK): Supply your own OpenAI, Anthropic, or Ollama API key. Keys are encrypted with AES-256 and stored locally — never sent anywhere except your configured LLM provider.

Domains Covered

• LLM Internals (transformers, attention, KV cache, quantization, speculative decoding)
• Retrieval-Augmented Generation (vector DBs, chunking, reranking, RAGAS evaluation)
• Fine-tuning & Alignment (LoRA, QLoRA, RLHF, DPO, dataset curation)
• Prompt Engineering (CoT, structured output, adversarial prompting)
• Agentic Systems & Tool Use (MCP 2026-07-28, ReAct, multi-agent, memory architectures)
• MLOps for LLMs (vLLM, TGI, observability, cost optimization)
• AI Safety & Ethics (hallucination mitigation, red-teaming, EU AI Act)
• LLM Evaluation & Quality (golden sets, LLM-as-judge, regression evals, online metrics)
• Security & Guardrails (OWASP LLM risks, injection defenses, output filters, tool isolation)
• Fault Tolerance & Reliability (retries, circuit breakers, failover, graceful degradation)
• Production GenAI Applications (cost, caching, streaming, observability, structured output)
• Reasoning Models & Test-Time Compute (o-series, R1, thinking budgets, distillation)
• Context Engineering (compaction, memory vs RAG, context rot, prefix cache)
• GenAI System Design (copilots, support agents, document QA, cost/latency/safety)
• Behavioral & Situational (STAR method, trade-off reasoning, production incidents)

Quick Start (< 5 minutes)

Prerequisites

• Node.js 18+ or Bun
• pnpm (npm install -g pnpm)
• An API key from OpenAI, Anthropic, or a local Ollama instance

1. Clone and install

git clone https://github.com/your-org/gen-ai-interview-prep.git
cd gen-ai-interview-prep
pnpm install

2. Set up the database and seed questions

cp .env.example .env
pnpm setup

This creates a local SQLite database and seeds it with 690+ questions. No API key needed for this step.

3. Start the app

pnpm dev

Open http://localhost:3000. On first launch you'll be prompted to add your API key.

Provider Setup

OpenAI:

• Get a key at platform.openai.com
• Select "OpenAI (GPT-4o)" in the setup screen
• Paste your sk-... key

Anthropic:

• Get a key at console.anthropic.com
• Select "Anthropic (Claude)" in the setup screen
• Paste your sk-ant-... key

Ollama (local, free):

• Install Ollama: curl -fsSL https://ollama.ai/install.sh | sh
• Pull a model: ollama pull llama3.2
• Select "Ollama / Custom" in setup — no API key needed

LM Studio (OpenAI-compatible local server):

• In LM Studio, start the local server and copy the server URL (for example http://192.168.1.5:1234).
• In setup/settings, choose "Ollama / Custom endpoint".
• Set Model to the LM Studio model id (for example qwen/qwen3.5-9b).
• Set Base URL to the LM Studio server URL. The app auto-appends /v1 when missing.
• API key is optional for local servers; a placeholder key is used if omitted.

Features

• Browse by domain with difficulty filter (Foundational / Intermediate / Advanced)
• Expand any question to see the full ideal answer with core concept, interview framing, key points, and follow-up questions
• Bookmark questions as "Studied" or "Needs Review"
• Full-text search across all questions and answers
• Random Question mode per domain for quick 10-minute reviews
• Add more questions (AI) per domain using your API key — new questions are appended to the bank, never replacing existing ones
• JSON export of the full question bank for contributor workflows

Security

• API keys are encrypted with AES-256-GCM using a randomly generated local key stored in .keyfile
• The .keyfile and database (local.db) are gitignored and never committed
• No telemetry, no analytics, no external requests except user-initiated LLM calls
• Network deployments: If the app is reachable beyond your machine, set APP_ACCESS_PASSWORD in the environment. Users must sign in at /login; the session is an httpOnly signed cookie (no bearer token in the browser bundle). Optionally set APP_SESSION_SECRET to a long random string so session verification avoids a slow key derivation on every request. Failed login attempts are rate-limited in memory per client IP (trust X-Forwarded-For only from your proxy). For maximum isolation, put the app behind a reverse proxy with SSO or Basic Auth instead.

Contributing

Question contributions are welcome! See CONTRIBUTING.md for how to add questions via JSON or submit a PR.

To export the current question bank:

# Via the UI: Domains page → Export JSON
# Or directly:
curl http://localhost:3000/api/export > questions-export.json

License

MIT — free to use, modify, and distribute.