SynthNet

DESCRIPTION:

SynthNet is a decentralized synthetic data marketplace built as a Bittensor subnet. 
AI teams today are hitting a wall  real training data is scarce, privacy-restricted, 
or locked behind six-figure licensing deals. Centralized providers like Scale AI and 
Gretel offer no transparency, no verifiable quality, and complete vendor lock-in.

SynthNet solves this with a three-sided protocol: Users request datasets specifying 
type, volume, format, and quality parameters. Miners across the Bittensor subnet 
compete to produce the best matching output. Validators score quality across five axes 
 spec adherence, diversity, accuracy, coherence, and format — and write those scores 
on-chain, immutably. TAO rewards flow automatically to quality producers. No central 
operator. No gatekeeping.

The core innovation is the incentive design: a multi-signal quality score weighted 
45% downstream utility (hidden benchmarks), 25% human audit, 20% automated checks, 
and 10% novelty via embedding distance. Concave reward aggregation (top-K scoring per 
epoch) makes spamming structurally impossible — volume cannot beat quality. Validators 
are scored against audit anchors, so collusion surfaces immediately and inaccurate 
scorers lose influence and dividends.

The result: the first open, permissionless, on-chain-verified synthetic data 
marketplace infrastructure for the entire AI data economy.

shipped a full-stack, production-ready frontend from scratch during the hackathon:

→ Built a multi-step dataset request flow with live TAO cost estimation, 
  type/spec/format configuration, and animated step transitions using Framer Motion

→ Built an interactive Dashboard with real-time network health stats (miners, 
  datasets, quality scores, uptime), quality line charts, radar charts per miner, 
  and a live generation queue

→ Built a Marketplace with dataset browsing, type/quality filtering, inline code 
  preview snippets, and a Download flow with toast confirmation

→ Built a Miner Explorer with on-chain participant data (UID, stake, trust, 
  consensus), 3D network topology visualization, and per-miner detail panels

→ Built an interactive 3D hero scene in Three.js / React Three Fiber — draggable, 
  zoomable, pannable network graph representing the decentralized subnet — with 
  animated node hover details and smooth pointer-event handling

→ Integrated a Connect Wallet flow with visual feedback

→ Designed and implemented the full incentive architecture: multi-signal quality 
  score formula, concave reward aggregation, validator accountability model, and 
  anti-gaming mechanisms (rotating benchmarks, canary strings, multi-validator 
  redundancy ≥3 per task)

→ Wrote full technical spec for subnet economics: dataset purchase fees, emission 
  pool distribution, and Enterprise API tier

Stack: React 18, TypeScript, Vite, Tailwind CSS, Framer Motion, Three.js, 
React Three Fiber, Recharts. All production-ready — mock data swaps to live 
Bittensor API calls on subnet deployment.

We havnt recieved any funding as of now