Vitalii Radionov@vitalii_radionov211
2
Events attended
2
Submissions made
1 year of experience
About me
I am an AI-Native Founder and Data Architect. I build automated trust intelligence systems and forensic agents that analyze on-chain data and corporate sentiment. Passionate about RAG, BigQuery, and using Agentic AI to solve massive security challenges in Web3 and Corporate spaces. Currently building Vartovii β a Zero-Trust Oracle.
π€ Latest Submissions
ProofMesh Treasury Guard
AI agents can trigger payouts quickly, but they still lack a machine-verifiable trust layer at payout time. ProofMesh Treasury Guard solves that gap by making trust itself programmable. Before treasury releases capital, the agent buys three live proofs from real external providers: WhoisXML for domain age and registrant evidence, GLEIF for legal-entity registry evidence, and GoPlus for wallet-risk evidence. Those proofs are aggregated into a trust bundle, converted into a permit artifact, and translated into a treasury action: RELEASE, CAP, HOLD, or BLOCK. The product is not just a demo dashboard. ProofMesh also exposes a Treasury Guard API, so other systems can issue and verify guard packages before allowing capital to move. In the current build, the live /demo surface shows the full agent loop, the /ledger surface shows inspectable receipts and settlement evidence, and the system already demonstrates a persisted 52-transaction Arc testnet story with sub-cent action-level transport economics. Most projects show that agents can pay. ProofMesh shows that agents can buy trust before paying, then expose that trust decision as a programmable control layer for treasury and AP workflows. That is the core thesis: buy trust before capital moves.
26 Apr 2026
Vartovii Sentinel-8004
`Vartovii Sentinel-8004` is a trust and control layer for autonomous trading agents. Instead of building another trading bot, we focused on the checkpoint that should exist before capital moves. Each trade starts as a canonical `TradeIntent`. Sentinel evaluates that intent against deterministic risk policy and returns one of three machine-readable outcomes: `ALLOW`, `DENY`, or `ALLOW_WITH_DOWNSIZE`. Every decision produces a signed verdict, validation artifact, permit verification result, and Kraken-shaped execution preview so judges can inspect the full proof chain end to end. The public repo implements real EIP-712 typed-data signing and verification, ERC-8004-style identity binding, organizer-aligned shared Sepolia contract configuration, and a hosted judge-first walkthrough with four canonical scenarios. The main demo surface is intentionally narrow: judges can open `/judge` for the proof story or `/operator` for a dry-run pipeline without needing private infrastructure, exchange credentials, or hidden services. This is the primary judged product surface. A separate founder-run companion repository exists only as supporting proof that the guardrail also matters in a live-market loop. As of April 8, 2026, that founder-run companion log contains 130 recorded cycles, including 83 approved shared-Sepolia RiskRouter actions. The public submission does not depend on that companion surface to run. Sentinel-8004 fits the hackathon thesis because it combines AI-agent execution intent, cryptographic verification, risk controls, and transparent auditability into one reusable control layer. The long-term product path is a deployable guardrail module or micro-SaaS for autonomous trading systems, treasury agents, and policy-constrained execution workflows.
12 Apr 2026
