Scale AI (today’s MVP) — Founder Kit

Overview

• Users: Project Manager, Annotator, QA Reviewer, Admin
• Platforms: Web app + lightweight Admin
• Data: image/text datasets, tasks, labels, reviews, exports
• Payments: usage-based SaaS (by task or dataset size)

MVP Scope (must-have)

• Auth (email + magic link), roles & permissions
• Project setup: import dataset, define label schema, create tasks
• Task routing: assign to annotators; simple work queue
• QA loop: approve/reject with comments; requeue on reject
• Analytics: progress %, per-label distribution, reject rate
• Export: JSON/CSV w/ task ids, label geometry/attrs

Core Screens

1. Projects list → create new project (schema, import)
2. Labeling workspace (keyboard shortcuts; next/prev task)
3. QA review queue (diff view, comments)
4. Analytics dashboard (throughput, quality)
5. Exports (format, sample, download)

Data Model (simplified)

User{id, email, role}
Project{id, name, ownerId, schema}
Task{id, projectId, inputUri, status: NEW|IN_PROGRESS|DONE|REVIEW|REJECTED}
Annotation{id, taskId, type, data, createdBy}
Review{id, taskId, decision: APPROVE|REJECT, comment, reviewerId}
Export{id, projectId, format, createdAt, url}
      

API (stub)

POST /api/projects { name, schema }
POST /api/projects/:id/import { files[] }
GET  /api/tasks?projectId=...&status=NEW
POST /api/tasks/:id/annotate { annotation }
POST /api/tasks/:id/review { decision, comment }
GET  /api/projects/:id/analytics
POST /api/projects/:id/export { format }
      

Go-to-Market (0–6 weeks)

• Target teams labeling images/text for GenAI fine-tuning
• Offer “first 5k tasks free” to seed datasets & testimonials
• Partner with agencies/collectives for overflow capacity

Risks & Mitigations

• Quality: add double-pass + consensus on high-risk tasks
• Throughput: queue autoscaling; hotkeys; pre-labels
• Security: signed URLs, per-project ACLs, audit log

Next Actions

• Ship clickable labeling workspace demo
• Import 1 public dataset; publish export samples
• Pilot with 2 design partners; iterate weekly