ReproLearn - Research Studio

Inspiration

Research papers often preserve the conclusion but lose the exact path that makes the result checkable: the source revision, permitted data, environment, command, expected observable, and definition of success. That gap makes independent reproduction expensive for researchers and opaque for students.

What it does

ReproLearn is a local-first Research Studio that turns one bounded paper result into a portable Research Capsule. An author reviews the claim, source, data boundary, license, sensitivity, execution command, and validator before sealing a checksum-bound contract. A constrained runner executes the approved scope. An independent receiver imports the capsule from a separate checkout, inspects its capability and data boundaries, reruns the same scope, and records match, divergence, skip, or failure as a receiver-owned receipt. The same capsule becomes a learning path with prediction, execution, explanation, and transfer.

How we built it

The UI is React and TypeScript, backed by a local loopback API and deterministic fixtures. Docker provides the bounded execution boundary with denied network access, read-only mounts, and resource limits. OpenAI assistance is optional and proposal-only; it can suggest structured claim or metadata proposals, while humans approve and the receiver owns verification. A green exit code is never presented as scientific truth.

Why it matters

ReproLearn makes research rerunnable, auditable, teachable, and reusable. Codex with GPT-5.6 accelerated the build by helping turn the product idea into an executable workflow, implement the control-plane UI and local API, generate focused tests, diagnose edge cases, and harden the release. The final engineering decisions preserve explicit human approval and evidence boundaries instead of hiding uncertainty behind AI.

Built With

• docker
• openai
• react
• typescript