The AI-native system of record for how a lab actually works.
A polished, clickable prototype of a JoVE product concept: a Lab Operating System that captures a lab's tacit know-how into a living Lab Knowledge Graph, auto-generates personalized onboarding from it, and verifies competence before anyone touches the bench.
This repository is an internal product concept and an unofficial prototype, not an official JoVE product, feature, or release. It contains no real JoVE assets, customer data, or content, and the "AI" is scripted, not a real model. The lab seed is built from real, publicly available information about one real lab (the McClean Lab at UW-Madison, via its public website and OpenWetWare wiki); a small set of non-public fields (the per-person training matrix, joined dates, and the departing/new-hire timing) is clearly marked as illustrative demo data, since capturing exactly those fields is the product's own value proposition. See the integrity boundary at the top of
src/lib/data.ts. This is NDA-sensitive and not for external distribution.
Live demo: https://labos-prototype.vercel.app
It is built to be shown to stakeholders (scientists and executives) instead of a written PRD. Every flow is set on a single cohesive real lab, the McClean Lab (biological signal processing in yeast, Dept. of Biomedical Engineering, UW-Madison), populated from that lab's public website and OpenWetWare wiki, so the data stays consistent end to end. Non-public fields are marked as demo (see src/lib/data.ts).
Four interactive flows, plus an 11-step guided tour that walks all four in story order with a spotlight overlay. A persistent left sidebar lets you jump between flows freely.
Connect a lab's existing docs, watch the AI extraction animation, and see the Lab Knowledge Graph assemble itself (force-directed, with clickable nodes). An AI interview captures the unwritten rules, and a personalized onboarding journey is generated from the graph. This is the "build once" idea.
A personalized path for a new rotation student surfaces the "why" behind each protocol step, then a competency checkpoint requires evidence. Switch to the trainer view to e-sign, and the status flips to Verified. Competence is proven, not just "watched."
A grounded AI co-pilot that answers with citation chips tied to the lab's own sources, and a safety guardrail that cites or abstains. It refuses to invent a hazardous step and honestly declines questions that fall outside its knowledge.
A people-by-skills competency dashboard that flags single points of failure (bus-factor-1 risk), benchmarks against the network, and runs a tacit-capture flow that structures a departing expert's know-how into the graph. Capturing one expert drops the bus-factor count.
The thesis behind LabOS is that the durable advantage is software IP, not video. The moat is a per-lab Lab Knowledge Graph (the system of record for how a specific lab actually operates), the AI layer on top of it, a Competency-Verification Engine that turns "watched a video" into auditable, signed-off proof of skill, and a cross-lab data flywheel that compounds as more labs come online.
The full product design (architecture, the AI IP layer, personas, flows, roadmap, and pricing) lives in the private connecting-dots repository at reports/jove-labs-product-design.md.
- Next.js 15 (App Router) and React 19 with TypeScript
- Tailwind CSS, with a single design-tokens file (
tailwind.config.ts) holding the WCAG-AA-corrected JoVE palette, radii, and shadows - Geist and Geist Mono, self-hosted via
next/font - Radix UI primitives (dialog, dropdown menu, tooltip) for accessible interactions
- react-force-graph-2d for the interactive Lab Knowledge Graph (client-only, loaded via
next/dynamicwithssr: false) - No backend, no database, and no real LLM. All state is React and in-memory, and the "AI" is scripted with realistic latency. Deployed on Vercel.
Requires Node.js 18.18 or newer.
npm install
npm run devThen open http://localhost:3000.
For a production build (also what Vercel runs):
npm run build
npm startlabos-prototype/
├─ src/
│ ├─ app/ # Next.js App Router: landing, layout, and /flows/* routes
│ ├─ components/ # brand, landing, shell, graph, tour, ui, and the four flows
│ └─ lib/ # real-lab seed (+ marked demo fields), graph builder, scripted "AI", types
├─ _screenshots/ # full-page captures of every flow, the landing, and the tour
├─ DESIGN-SYSTEM.md # design tokens, color, type, and component conventions
├─ tailwind.config.ts # the JoVE design-token source of truth
└─ next.config.mjs
The visual language (tokens, the WCAG-AA JoVE palette, typography, and component conventions) is documented in DESIGN-SYSTEM.md. WCAG 2.1 AA is a stated JoVE brand requirement and is implemented throughout: a skip link, semantic landmarks, correct heading order, keyboard navigation, visible focus rings, ARIA labels, and AA-contrast color.
This is an early-stage prototype built to communicate a product concept. The lab seed is real, publicly sourced data about the McClean Lab (with non-public fields clearly marked as illustrative demo); it uses no real JoVE assets and no real AI. It is proprietary and confidential, NDA-sensitive, and not for external distribution.
All rights reserved. This is not open-source software and ships with no license grant.