The problem Laytus solves

The Problem
Current platforms like Polymarket and Kalshi remain stuck in a binary primitive stage. They function as simple marketplaces for single-event outcomes, while traditional gambling markets internalize risk instead of outsourcing flow to market makers. In mature sports betting markets (2024–2025 data from NY/IL/NJ), parlays consistently drive over 70% of total operator revenue despite being a fraction of the total volume. The house edge on a single bet is ~5%, but on multi-leg parlays, it compounds to 15–25%. By only supporting binary, single-event markets, current DeFi prediction platforms are ignoring the most profitable segment of the industry. Additionally, in DeFi, liquidity is fragmented across thousands of isolated AMM pools. LPs are forced into passive, low-yield positions rather than capturing the high-margin vig (house edge) from sophisticated, multi-leg products.

Laytus is the first Unified Liquidity Vault on ADI designed specifically to underwrite the complex prediction economy. We operate through the following mechanisms:

The unified house vault: LPs deposit USDC into a single pool that acts as the counterparty for all thematic stacks. Instead of fragmented liquidity, we aggregate capital to absorb multi-leg risk.

The first on-chain Compound Vig products: By allowing users to bundle bets, the vault captures the 15–20% house hold that traditional sportsbooks use to dominate the market.

AI-Risk Layer (Off-Chain/On-Chain Hybrid): Our off-chain engine models real-time correlations and issues Risk Certificates (EIP-712 signed payloads). These are validated by 10 on-chain safety guardrails (e.g., Exposure Caps, Delta-Neutral Limits) to ensure the vault never over-leverages on a single outcome, while also taking into account factors like duration and correlation risk to maximize profitability for vault depositors.

ADI Native Execution: Laytus is built on ADI to provide sub-cent gas fees and millisecond response times required to compete with centralized sportsbooks while maintaining on-chain transparency.

Challenges we ran into

Challenges we faced

1. Testnet Liquidity & Access Bottlenecks
   We encountered significant friction in sourcing enough Testnet tokens due to strict social verification requirements (X/Twitter follower minimums and other proof of humanity measures). Additionally, simulating on chain, high-frequency testing proved challenging given current faucet limits. We solved this by optimizing our gas usage and batching transactions to preserve our limited testnet resources and running off chain monte carlo simulations.

2. Architectural Pivot: Transitioning to ADI
   Moving the project to ADI (Asset-Driven Infrastructure) after our initial development on Base required a rethink of our asset settlement logic. We had to ensure our AI Risk Engine remained chain-agnostic while maintaining the sub-second latency required for real-time quoting.

3. AI Model Training & Data Normalization
   Training a model on live Polymarket data presented unique challenges, particularly in normalizing event-driven signals that are often noisy or highly volatile. Implementing the Sentence-Transformers stack allowed us to better correlate social sentiment with price action, significantly improving our model's predictive reliability.

Use of AI tools and agents

Use of AI Tools and Agents

The Risk Architect agent runs as a headless service to identify non-obvious market correlations and price our products in real-time. It constantly ingests Polymarket order books to find links between events like Fed rate hikes and ETH price volatility. Using Gaussian copula math and a dataset of hundreds of thousands of polymarket trades we trained the model on and plan to retrain on biweekly polymarket data, it autonomously generates and signs EIP-712 Risk Certificates, ensuring every multi-leg stack is mathematically sound before it ever reaches the user.
We deploy ephemeral Sentinel Guardian pods for every high-stakes transaction to act as a private, isolated risk auditor. When a user builds a complex parlay, a dedicated pod spins up to run a 10-step ML pipeline that simulates how black swan scenarios would impact the protocol's solvency. By destroying these pods immediately after the session, we maintain total data isolation between institutional users and prevent any single ticket from creating systemic risk for the LP Vault.
Our Sovereign Liquidity Nodes combine local Qwen 2.5 models with native ADI Chain nodes to manage the treasury without human intervention. These nodes run an autonomous loop that identifies liquidity gaps in global markets, updates the protocol’s risk map on the 0G Data Availability layer, and allows users to analyze and maintain their risk. If the house exposure hits a predefined threshold, the protocol will no longer allocate that specific product to users
We utilized claude code to build and refactor the core backend microservices, and codex 5.3 for our front end. We ran Claude in agentic mode to handle the heavy lifting of our Dockeriz