Praxis

Inspiration

While AI has revolutionized the way we interact with information, it remains largely excluded from high-stakes financial execution due to one fatal flaw: Probabilistic Hallucinations.

In most current AI-to-Blockchain implementations, an LLM’s output is trusted blindly. If the AI "guesses" a wrong amount or an invalid date, capital is lost forever. We were inspired to build a system where the "intelligence" of AI is protected by the "certainty" of digital physics. We wanted to solve the "Hallucinating Banker" problem by building a Neurosymbolic bridge to the XRP Ledger.

What it does

Praxis is a deterministic AI-to-XRPL gateway. It allows users to manage institutional assets (like RLUSD) using natural language, but intercepts the AI’s probabilistic intent with two layers of hard mathematical logic:

Neural Parsing: Gemini 3.1 Pro interprets messy human requests into structured transaction drafts. Formal Verification (Flux Enclave): The system compiles the AI’s draft into a custom bytecode language and runs formal proofs to ensure the transaction obeys absolute safety laws (e.g., amount bounds, temporal logic). Game-Theoretic Scheduling (Tenet Engine): A Nash-Equilibrium engine that strategicially prioritizes transactions in the queue based on economic utility, ensuring that critical high-value capital always has the most rational path to the ledger.

How we built it

We engineered a triple-layer Neurosymbolic stack:

The Frontend: A high-performance Next.js 14 dashboard using Server-Sent Events (SSE) to stream concurrent traces of "Neural" and "Symbolic" logic side-by-side. The Brain: Google’s latest Gemini 3.1 Pro API, used to power the Layer 1 intent engine. The Enclave: A custom C99-based verification engine (Flux) that provides deterministic state-machine validation before any asset is moved. The Solver: A strategic layer built in the Tenet DSL, modeling the transaction queue as a game theory problem solved via a Java-orchestrated Nash solver. The Ledger: Direct integration with the XRP Ledger (XRPL) using xrpl.js for cryptographic signing and submission of on-chain escrows.

Challenges we ran into

Temporal Epoch Syncing: We faced a significant bug where traditional unixTime was generating negative Ripple Epoch offsets for 2026 dates. We had to deep-dive into the XRPL timestamp protocol to implement a corrected isoTimeToRippleTime conversion. Native Module Infrastructure: Integrating high-speed WebSocket buffers in a serverless environment required specialized configuration to handle the bufferutil and utf-8-validate native dependencies. Disk Bottlenecks: High-velocity development and browser tracing consumed our root partition (100% full), forcing us to build a custom cleanup routine to sustain the build environment.

Accomplishments that we're proud of

Atomic AI Safety: We successfully built a system that "kills" a transaction if the AI hallucinates even a single byte of data. Game Theory Integration: Successfully solving for Nash Equilibrium in real-time to decide transaction priority is a level of sophistication rarely seen in DeFi gateways. End-to-End Reliability: Moving from a natural language sentence to a confirmed transaction hash on the public XRPL testnet in under 60 seconds with full formal proofs.

What we learned

We learned that the future of AI in finance isn't "better prompts"—it's better constraints. By building "Neurosymbolic" systems, we discovered that you can utilize the most creative AI models (like Gemini 3.1) without inheriting their risks, as long as you have a deterministic "Symbolic" layer to act as the firewall.

What's next for Praxis

Multi-Escrow Batching: Allowing atomic deployment of multiple related financial intents in a single ledger pulse. Governance Integration: Allowing DAOs to set custom "Flux Scripts" to enforce their own unique governing laws on the gateway. Mainnet Transition: Moving from XRPL Testnet to high-value Mainnet execution for institutional RLUSD management.

Built With

• c99
• flux
• github
• javascript
• name.com
• next.js
• ripple
• rlusd
• tenet