VN Vu Nguyen Airwallex SG proof packet Resume
AI-native financial systems

Finance for agents, with receipts.

Production data agents at Phantom. Private payments on Tempo. Agent-commerce contracts and proof verification on Arbitrum. The interesting part is not generating code faster; it is using agents while keeping money, permissions, and claims bounded by systems you can inspect.

LOCATIONSingapore-ready
CONTRACTSDeployed
STATUS
Transaction trace
inspectable agent-finance paths
ONCHAIN RECEIPTS
NetworkTempo Moderato
PrimitiveShielded pool
Result3 confirmed txs
Each step opens its Tempo receipt. Demo verifier; production arkworks prover remains private.
Pull on any thread

Ask the application

The resume compresses nine years into two pages. This answers against the underlying work and points back to inspectable evidence.

Grounded Q&A

Interrogate the fit, not the pitch.

Ask about production agents, financial-system boundaries, data-platform depth, project choices, or the move to Singapore.

Public-safe profile corpus
Evidence links returned with answers
Vu / evidence assistant local fact set

Grounded in public-safe facts. Hosted answers use a small model; this file falls back to embedded retrieval over the same evidence.

Receipts, not claims

Inspect the live evidence

The resume is the index. These are the things a technical evaluator can actually open: product, source, deployment, and execution.

One agent-commerce path

Intent to settlement

Open visual explainer
01 / DISCOVER

Agent finds merchant

ERC-8004 registry returns deal catalog, endpoints, and stealth meta-address.

02 / PLAN

Pure recipe

One planner creates calldata with no side effects or network calls.

03 / PAY

Private execution

Stealth mode hides the recipient; Railgun mode hides sender and recipient.

04 / RECEIPT

Encrypted evidence

XChaCha20-Poly1305 receipt is readable only by the buyer agent.

05 / PROVE

Anonymous loyalty

Merchant verifies spend threshold without identity or purchase history.

The operating history underneath it

AI-native, not AI-decorated

Agents are part of the implementation process and the product surface. Contracts, evals, sandboxes, and money-path boundaries decide where they are allowed to act.

Production agent / Phantom

Warehouse exploration agent

Sandboxed Python analysis, MCP surface, held-out retrieval evals, per-turn traces, SQL validation, and bounded tools.

15+ MCP tools · recall@k / MRR
Financial system / Rubick

Non-custodial AI trading

LLM-classified market evidence feeds a durable execution system whose analytics side cannot hold keys or trigger trades.

14-state saga · ~1,580 commits
Agent tooling / Lum

Parallel-agent supervision

Spatial threads, microVM isolation, structured agent state, permission visibility, and multi-tier pause/resume recovery.

QEMU/HVF · Firecracker · Tauri/Rust