-
-
-
-
-
-
-
-
-
-
-
-
HELP AND SUPPORT CENTRE
-
HELP SUPPORT
-
DASBOARD
-
To test the demo, log in with: Email: [email protected] | Password: Password123!
Inspiration
The era of Agentic Commerce has arrived. Autonomous AI agents can now negotiate, purchase, and initiate transactions. However, traditional financial rails were built for a human-only world. Giving an autonomous agent direct access to corporate funds invites existential risks—from prompt injections and model hijacking to hallucination-induced capital loss. We asked ourselves: Who controls the money when AI takes action? This inspired us to build JOBENPAY, the mission-critical Trust Layer between autonomous AI agents and enterprise financial networks.
What it does
JOBENPAY enforces a strict architectural Golden Path: AI Proposes ➔ JOBENPAY Verifies ➔ Humans Authorize ➔ PayPal Disburses. Instead of giving AI direct access to a wallet, JOBENPAY intercepts the transaction and subjects it to:
Deterministic Mandates: Strict rules on budget, velocity, and time.
Decision Integrity Proof (DIP): Using Gemini 3.8 Flash to score the semantic alignment of the AI's reasoning against the human's original intent. AI can only tighten security, never loosen it.
Human-in-the-Loop: Dual-custody approval queues with cryptographic timestamps.
Immutable Ledger: Every approved transaction is secured using Post-Quantum Cryptography (ML-DSA-65) and Ed25519 signatures, creating a tamper-evident hash chain.
How we built it
We built the core engine using Node.js 22 LTS, React 18, and PostgreSQL for multi-tenant Row-Level Security.
AI Engine: We utilized Gemini 3.8 Flash as our intelligence layer for deterministic intent extraction and semantic reasoning.
Payments: Integrated directly with the PayPal Sandbox API using OAuth 2.0 and Payouts API. We ensured multi-level idempotency to prevent duplicate executions.
The Ledger: We implemented a strict Double-Entry Ledger Architecture enforced by PostgreSQL triggers to ensure zero floating-point errors.
Security First: We developed a Zero-Leak Environment ensuring absolute isolation between Sandbox and Live modes, backed by over 1,132 automated tests.
Challenges we ran into
Monotonic AI Logic: Engineering the AI so that it is architecturally impossible for it to downgrade a deterministic rejection into an approval. We had to treat AI output strictly as untrusted data.
Double-Entry Integrity at Scale: Ensuring that every single micro-transaction generated strictly balanced debit/credit legs without failing under high-velocity simulation.
Race Conditions: Preventing duplicate disbursements when network drops occur during PayPal API calls, which we solved by implementing tenant-scoped Idempotency Keys.
Accomplishments that we're proud of
1,132 Automated Tests Passed: Achieving a 100% green exit code across unit, integration, concurrency, and adversarial prompt-injection tests.
Post-Quantum Security: Implementing ML-DSA-65 keys, securing the platform against future quantum decryption threats.
Seamless Human-AI Governance: Creating a UI where users can visually see the AI's reasoning trace and mathematically verify the cryptographic hash chain with a single click.
What we learned
We learned that building for Agentic Commerce requires a fundamental shift in mindset. AI should not be treated as a trusted employee, but as a highly capable yet unpredictable entity that requires strict Deterministic Policy Gates. We also deepened our expertise in Post-Quantum Cryptography and the nuances of PayPal's Payout APIs
What's next for JOBENPAY
Live Production Rails: Moving beyond the PayPal Sandbox to real-world enterprise disbursements.
MCP Integration: Expanding our Model Context Protocol (MCP) server to allow any AI framework (Claude, AutoGPT, Cursor) to plug into JOBENPAY natively.
Decentralized Trust Registry: Integrating with vLEI (verifiable Legal Entity Identifiers) to create a global, decentralized ledger of trusted AI agents.
Built With
- docker
- node.js
- paypal
- post-quantum-cryptography
- postgresql
- react