Skip to content

Atrium

Atrium is an agentic technical interview platform for the AI era.

Core trick

Streams a candidate's sandboxed Claude Code session to an observer agent that scores behavior against the interviewer's custom rubric in real time.

Tags

Description

Atrium is an agentic technical interview platform for the AI era. LeetCode is dead. Candidates work with AI tools and debug real systems, so why test them on binary tree inversions? Atrium replaces coding puzzles with paid, real-world challenges powered by three Opus 4.6 agents. The Interview Architect ingests a JD, resume, and SDK docs to generate a realistic challenge using the company's actual stack, plus a custom weighted rubric. Candidates work in a cloud sandbox with Claude Code as an AI assistant that guides rather than solves. The Session Observer detects behavioral phase shifts and flags signals against the rubric in real-time. The Interviewer Copilot suggests follow-up questions tied to stated priorities. A comprehensive scorecard closes the loop. Candidates get paid. Interviewers get signal, not theater. Try it: https://atrium.fly.dev FAQs: https://docs.google.com/document/d/1z8_ud46tWJJLlCgysqLNhmFvkMTkYbFyfdCpRCAExZs
suraj-phanindra/Atrium★ 00 forksTypeScriptCreated Feb 14, 2026 · last push Feb 16, 2026

README excerpt

# Atrium

**See how they think.**

Atrium is an AI-powered technical interview platform that replaces LeetCode with paid, real-world challenges. Interviewers set up custom interviews by chatting with an Opus 4.6 agent — uploading a job description, candidate resume, and their SDK documentation. The AI generates a tailored coding challenge and builds a custom evaluation rubric.

During the session, candidates work in a cloud sandbox (E2B) with Claude Code. Three Opus 4.6 agents work together: an Interview Architect that designs the challenge through conversation, a Session Observer that evaluates the candidate's reasoning patterns against the custom rubric in real-time, and an Interviewer Copilot that suggests follow-up questions based on the interviewer's stated priorities.

The interviewer's dashboard shows a live swim-lane timeline of the candidate's activity — AI prompts, file changes, terminal commands, reasoning signals — with an AI-generated evaluation scored against their custom rubric.

**Live:** [atrium.fly.dev](https://atrium.fly.dev/)

**Built for the "Built with Opus 4.6" Claude Code Hackathon (Feb 2026).**

Team