Skip to content

Team HCI

I'm an HCI researcher, and I keep running into the same problem: every claim has to be tested with real users.

Core trick

Runs a simulated pilot with twelve composite personas and a three-reviewer panel, then audits reviewer disagreement in a second model pass.

Tags

Description

I'm an HCI researcher, and I keep running into the same problem: every claim has to be tested with real users. Two to three months of recruitment per study, another six months for the publication cycle, and very little room to discover — halfway through — that something basic about the design was off. Probe is a rehearsal stage for study design, powered by Claude Opus 4.7. You start with a one-sentence premise. Opus splits it into three sub-research-questions — mechanism, intervention, lived experience. Sonnet 4.6 surveys arXiv, ACM, and Semantic Scholar per angle. Opus then drafts three candidate study designs; you pick one, and Probe writes the actual artifacts: implementation plan, IRB memo, validation protocol, survey kit, diary kit. A simulated pilot runs twelve composite personas through the protocol — surfacing failure modes before recruitment. The final stage is an adversarial peer-review panel; Opus 4.7 then audits their disagreement and separates real conflicts from apparent ones.
Apolotary/probe-researcher★ 10 forksHTMLApache-2.0Created Apr 21, 2026 · last push May 1, 2026

README excerpt

<p align="center">
  <img src="./assets/probe_wordmark.svg" alt="probe — a rehearsal stage for HCI study design" width="640"/>
</p>

> **A rehearsal stage for HCI study design.** Type a research premise. Probe walks it through seven Claude-driven stages — interrogator → literature → methodology → artifacts → simulated evaluation → report → simulated peer review — and walks out with a sharpened study, draftable artifacts, and a panel of three reviewers who genuinely disagree about whether the study should exist. Then **Opus 4.7** audits the disagreement and tells you which conflicts the area chair must NOT average away.
>
> *Rehearsal stage for research. The performance still needs humans.*

Built solo for the [Cerebral Valley **Built with Opus 4.7** hackathon](https://cerebralvalley.ai/e/built-with-4-7-hackathon), April 21–26 2026.

---

## For judges: fastest path (≤90 seconds, no API key)

```bash
git clone https://github.com/Apolotary/probe-researcher.git
cd probe-researcher && npm install && npm run build
npx probe ui --web                # opens http://127.0.0.1:4470/ui
```

1. Click **`▶ replay sample run`** in the left sidebar.
2. Pick **`focus-rituals`** — the workflow opens with the bundled demo's premise pre-filled.
3. Walk through the seven stages (brainstorm → review). Each is paced for inspection; total replay ≈14 s.
4. **At the brainstorm stage**, select two RQ cards → click `merge A + B` → watch Opus 4.7 synthesize a hybrid sub-question.
5. **At the review stag

Team