Skip to content

Centient

Train AI, cent by cent

Category: Product & client work

Hackathon Project

Visit the live site(opens in a new tab)

  • Built with:
  • Next.js 16
  • React 19
  • TypeScript
  • Tailwind CSS 4
  • PostgreSQL
  • Prisma 7
  • Stellar SDK
  • Freighter
  • USDC on Stellar
  • Node.js workers
  • Redis
  • JWT/JOSE
  • Resend
  • Recharts
  • Sentry
  • PostHog
  • Vitest
  • Docker
  • Railway
The Centient core loop on mobile — one prompt, two AI responses to compare, and the USDC micro-reward for choosing the better one.

What it is

A two-sided data-labeling marketplace that pays people in USDC on Stellar to teach AI what “better” looks like. A contributor compares two AI responses, picks the better one with a one-line reason, and earns a micro-reward (0.05 USDC); AI labs fund campaigns and export clean, quality-controlled datasets.

Value

  • AI labs get neutral human preference data — the RLHF fuel worth $1–10 a datapoint — with quality engineered in.
  • Contributors in emerging markets earn real dollars from a phone, cashable to local currency via Stellar anchors.
  • Stellar gains a non-speculative flow: net-new wallets, real USDC velocity, every payout verifiable on-ledger.

Technology highlights

  • Stellar micropayments, accumulate-then-withdraw — sub-cent fees make paying five cents at a time viable; earnings settle on-chain once at cash-out via a payout worker and reconciler.
  • Engineered quality and anti-fraud — hidden gold-standard tasks, inter-annotator agreement, reason-spam and bias guards, prepaid campaign budgets.

More projects