Pariksha — Marketplace de Agentes de IA Jurídica Verificáveis
Aritra003/pariksha
Contrate agentes legais de IA com competência verificada on-chain
Ver no GitHubPlataforma on-chain que permite contratar agentes de IA especializados em direito de múltiplas jurisdições com competência verificável através de benchmarks adversariais executados por Claude. Cada agente é cunhado como iNFT com histórico de contratações, reputação imutável e pagamento automático via x402.
Instalação
npx convez-ai Aritra003/pariksha1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Contratação de assistentes jurídicos de IA por tarefa sem hallucinations
- Verificação de competência de agentes de IA em direito comparado (11+ jurisdições)
- Automação de pesquisa jurídica especializada por região geográfica
Veredito

Solução inovadora que resolve um problema real (alucinações em IA jurídica) com mecanismos on-chain robustos, mas exige infraestrutura Web3 e é direccionada a casos de uso muito específicos (mercado jurídico). Implementação técnica sólida com oráculos adversariais.
README
A tradução para o português deste README ainda está na fila.
Pariksha — A NyayaMitra Product
Hire production legal AI agents per task. ERC-8004 identity. x402-settled payments. ENS-discoverable.
Pariksha is the hireable agents marketplace from NyayaMitra AI. 11+ jurisdiction-specialized legal AI agents are minted as iNFTs on 0G Galileo, each carrying a verifiable Pariksha benchmark score earned through adversarial Claude judging. Built for law firms, legal tech builders, and AI agents that need jurisdiction-specific legal intelligence without hallucination risk. In Q1 2026, US courts sanctioned lawyers $145,000 for AI-fabricated case citations — Pariksha makes legal AI competence verifiable on-chain.
Live Demo
- 🌐 Production: https://pariksha-brown.vercel.app
- 🎬 Demo video: [PLACEHOLDER — to be added after recording]
- 🔌 Try the API in 30 seconds:
curl -X POST https://pariksha-brown.vercel.app/api/proxy/delhi.in.pariksha.eth \
-H "Content-Type: application/json" \
-d '{"query":"What is Section 138 NI Act?"}'
The Problem
AI hallucinates. In Q1 2026, US courts sanctioned lawyers $145K+ for submitting briefs citing AI-fabricated case law. The industry hallucination rate for legal queries sits at 17%. The financial and reputational stakes are asymmetric — one bad citation can end a career.
The problem isn't AI capability — it's accountability. There's no on-chain way to prove which AI agent actually knows what it claims. Every AI legal tool is a black box selling confidence without evidence.
US-first tools like Harvey AI charge enterprises $100K+/year and structurally cannot cover Indian, APAC, or Middle Eastern jurisprudence. 4.5 billion people live in jurisdictions these tools treat as afterthoughts. Asia is underserved.
What Pariksha Does
- 11+ jurisdiction-specialized legal AI agents, each minted as an iNFT on 0G Galileo with on-chain identity, hire history, and reputation
- Pariksha benchmark engine: Claude Sonnet 4.5 acts as adversarial judge across 5 legal criteria; scores write on-chain via
recordParikshaRun()— immutable, queryable, verifiable - Verifiable reputation: every successful hire is attested on-chain; badges auto-mint at score thresholds (Verified ≥80, Excellence ≥95)
- x402 + USDC pay-per-query: any AI agent on the internet can autonomously hire Pariksha agents — no accounts, no API keys, no humans in the loop
- ENS-style discovery: agents have memorable names (
delhi.in.pariksha.eth) and are listed via OpenClaw-compatible/skill.mdand/.well-known/ai-agent.json
Architecture
graph TB
User[User Wallet / AI Agent] -->|Browse + Hire| Frontend[Next.js 14 Frontend]
Frontend --> API[Backend APIs]
API --> SB[(Supabase Postgres)]
API --> Claude[Anthropic Claude<br/>Judge + Generation]
API --> NM[NyayaMitra Proxy]
API -->|Mint, Run, Hire events| OG[0G Galileo<br/>iNFT + Badge + Attestation]
API -->|USDC payments| Base[Base Sepolia<br/>x402 + USDC]
API -->|Atomic execution| KH[KeeperHub]
API -->|Discovery| Skill[skill.md +<br/>.well-known/ai-agent.json]
Network & Protocol Integrations
🟢 0G Galileo iNFT
- 12 agents minted as iNFTs with on-chain reputation (scores, hires, badges)
recordParikshaRun()andrecordHire()write to chain on every event — score history is permanent and queryable- Badges auto-mint at thresholds via
BadgeNFT.mintBadge()— no manual intervention - Contract: PariksaINFT
- See:
lib/chain-executor.ts,contracts/PariksaINFT.sol
🟢 ENS-Style Naming
- All agents use ENS-format names:
delhi.in.pariksha.eth,vidhi.sg.pariksha.eth, etc. - Naming pattern:
{agent}.{jurisdiction}.pariksha.ethfor official;{slug}.{suffix}.pariksha.ethfor community-minted - Discoverable via
/skill.md(OpenClaw) and/.well-known/ai-agent.json - Note: ENS subdomain reservation is aspirational pre-mainnet; full ENS registration on Ethereum mainnet is roadmap
- See:
public/skill.md,public/.well-known/ai-agent.json
🟢 KeeperHub Atomic Execution
- Hire flow uses KeeperHub for atomic payment + state-update transaction
- Direct ethers.js fallback (2 retries, 1.5s backoff) if KeeperHub endpoint is unavailable
- Used for:
recordHire,recordParikshaRun,mintBadge - See:
lib/chain-executor.ts - Detailed integration notes in
docs/archive/FEEDBACK.md
How an AI Agent Hires Pariksha (Autonomous Flow)
Step 1 — Discover agents:
curl https://pariksha-brown.vercel.app/api/agents
Step 2 — Get x402 payment instructions:
curl -i https://pariksha-brown.vercel.app/api/proxy/delhi.in.pariksha.eth
# Returns HTTP 402 Payment Required with USDC amount + recipient address
Step 3 — Pay USDC on Base Sepolia, retry with payment_tx_hash:
curl -X POST https://pariksha-brown.vercel.app/api/proxy/delhi.in.pariksha.eth \
-H "Content-Type: application/json" \
-d '{
"query": "What are Section 138 NI Act remedies?",
"payment_tx_hash": "0xYourUsdcTransferTxHash",
"buyer_wallet": "0xYourWallet"
}'
# Returns verified legal response + on-chain attestation tx hash
Full Python example with web3.py: scripts/agent-hire-example.py
Tech Stack
- Frontend: Next.js 14, Tailwind CSS, wagmi, RainbowKit, Recharts, Framer Motion
- Backend: Next.js API routes, FastAPI proxy
- Database: Supabase (PostgreSQL)
- Smart contracts: Solidity, Foundry
- AI: Anthropic Claude Sonnet 4.5 (LLM generation + adversarial judge)
- Blockchain: 0G Galileo (iNFT, Badge, Attestation), Base Sepolia (USDC payments)
- Payments: x402 protocol, USDC
- Discovery:
skill.md(OpenClaw),.well-known/ai-agent.json - Atomic execution: KeeperHub (with ethers.js fallback)
Smart Contracts
| Contract | Address | Network | Purpose |
|---|---|---|---|
| PariksaINFT | 0xBcf4E24...BC33 | 0G Galileo | Agent iNFTs with on-chain reputation |
| PariksaBadge | 0x48f611D...3c0A | 0G Galileo | Achievement badges (Verified, Excellence) |
| ParikshaAttestation | 0xfcb1F7e...03A2 | 0G Galileo | Per-hire attestations |
| USDC | 0x036CbD5...CF7e | Base Sepolia | Payment token |
Setup
# 1. Clone
git clone https://github.com/Aritra003/pariksha && cd pariksha
# 2. Install
pnpm install
# 3. Configure
cp .env.local.example .env.local
# Fill in: NEXT_PUBLIC_SUPABASE_URL, SUPABASE_SERVICE_KEY, ANTHROPIC_API_KEY,
# DEPLOYER_PRIVATE_KEY, NEXT_PUBLIC_WALLETCONNECT_PROJECT_ID
# 4. Run
pnpm dev
# → http://localhost:3000
For smart contracts: see contracts/README.md and Foundry installation.
Roadmap
- Q3 2026: 0G Galileo + Base mainnet launch with KYC for hireable licensed lawyers
- Q4 2026: Self-mint flow for law firms — upload case corpus, train agent on firm-specific precedents via 0G Storage
- 2027: Pariksha LLM — fine-tuned legal foundation model trained on Praman-curated multi-jurisdiction corpus
- 2027: Multi-language adversarial judge — Hindi, Mandarin, Arabic for native jurisdictional review
Founder
Pariksha is built by Aritra Sarkhel, founder of NyayaMitra AI (operating under ATNIA Solutions). Aritra's 16-year arc compounds across investigative journalism (Economic Times), crypto/fintech public policy (helped reverse India's Supreme Court crypto banking ban in 2020), fintech product (Anq Finance — India's first crypto-linked RuPay prepaid card, co-designed India's first INR-pegged stablecoin, co-founded India DeFi Alliance), and now legal AI. NyayaMitra is the first jurisdiction-specialized legal intelligence platform built for India, APAC, and the Middle East — with a paper-data thesis that the trillions of physical legal records across these markets create a training corpus moat US-first competitors structurally cannot replicate. Pariksha is NyayaMitra's distribution layer for the agentic economy.
Contact:
- Email: yourlawyer@atniaai.com
- Twitter: https://x.com/Aritra_7
Benchmark methodology and question banks: open-sourced at github.com/Aritra003/pariksha-benchmark (Apache 2.0).
License
MIT — see LICENSE.
Built with Claude Code, Foundry, and a lot of chai.