Hire AI engineers who actually ship.

LunaPrompts is the AI-native technical screen. AI-native tests for prompts, agents, and RAG. Not algorithm puzzles. Founder-led pilots. 1,000+ engineers screened.

· Try a sample assessment

When AI solves the test in 5 seconds, you stop testing the test.

71% of hiring leaders say AI is making technical skills harder to assess. The legacy assessment stack (HackerRank, CodeSignal, Codility) was built for 2018 SWE hiring. LunaPrompts is the rebuild: prompt engineering, agent code, and RAG are first-class assessment types with their own grading pipelines.

What LunaPrompts tests

  • Prompt engineering — iterate against test cases with model picker (gpt-5.1, gpt-4.1, kimi-k2, llama-3.2, gemini-2.5-flash) and full hyperparameter control (temperature, top_p, max_tokens, frequency and presence penalty).
  • Agent code — multi-file Python projects in real Judge0 sandbox with tool-call loops and .env files.
  • Algorithms — 9 languages with public + hidden tests.
  • SQL — in-memory DuckDB on real datasets.
  • System design — visual canvas, rubric-graded.
  • Multiple choice — fast top-of-funnel filter.

AI isn't a tool we allow. It's the test itself.

Every challenge puts an LLM in the candidate's hands. They prompt, iterate, pick models, debug. We grade the craft, not the final answer. Anti-cheat is for impersonation, not AI tool use.

How a pilot works

  1. Day 0 — Pick a role. Send one open AI engineering req. Custom challenge mix assembled within 24 hours.
  2. Days 1–7 — Run candidates. 10 to 30 invites by CSV or one-by-one. Browser-based, auto-graded, with built-in proctoring (FaceMesh, fullscreen, tab-switch, clipboard, devtools blocking).
  3. Day 7 — 20-minute readout. Walk the scorecards together, side-by-side with your current screen.

No procurement cycle. No platform commitment. Book a pilot call.

Frequently asked questions

How is LunaPrompts different from HackerRank or CodeSignal?

HackerRank and CodeSignal bolt AI questions on top of 2018 software engineering assessments. LunaPrompts rebuilt the whole thing AI-native: prompt engineering, agent code, and RAG are first-class assessment types with their own grading pipelines.

Can candidates use ChatGPT during the test?

Yes. LunaPrompts hands candidates a model picker and hyperparameter controls inside the assessment. We grade iteration, judgment, and final signal — not whether they avoided AI.

Can we run LunaPrompts alongside our current screen?

Yes. Most pilots run LunaPrompts in parallel with your existing assessment platform so you can compare signal directly. No integration is required to start.

How do you prevent impersonation and outsourcing?

In-browser MediaPipe FaceMesh face detection, fullscreen enforcement, tab-switch detection, clipboard and devtools blocking, with auto-submit on configurable warning thresholds.

Does LunaPrompts integrate with ATS like Greenhouse, Lever, Ashby, or have SAML SSO?

CSV and XLSX candidate invites today. Greenhouse, Lever, Ashby integrations and SAML SSO are on the roadmap, prioritized by design partner request.

How does LunaPrompts pricing work?

Pilot first with no commitment, typically one role and 10 to 30 candidates. Annual plans are seat-based for the team running assessments, plus volume tiers for candidates evaluated.