Payments
ISO 20022, end to end: pain.001, pacs.008, camt.053, and acmt.001. Adopt one library at a time.
Explore the ISO 20022 librariesSenior Product Manager, Wholesale Payments & API · HSBC
I lead wholesale payments API products at HSBC and build open-source ISO 20022 and post-quantum tooling that banks and fintechs can inspect, test and reuse.
EPAA white-paper contributor · Fellow, Institution of Analysts and Programmers

Payments and digital products, built at:
# 1. Validate ISO 20022 pacs.008.001.10 (pip install pacs008)
pacs008 -t pacs.008.001.10 -m template.xml -s schema.xsd -d payments.csv --dry-run
✓ Schema validation passed
✓ Data validation passed (1 payment records)
# 2. Verify NIST FIPS 203 ML-KEM-768 KEM (cargo add kyberlib)
cargo test -p kyberlib --test test_kem
✓ test keypair_encap_decap ... ok
✓ test result: ok. 9 passed; 0 failed; finished in 0.01s
# 3. Verify live CycloneDX SBOM provenance (SLSA / Sigstore)
curl -sSLO https://sebastienrousseau.com/sbom.cdx.json && gh attestation verify sbom.cdx.json --owner sebastienrousseau
✓ Loaded 1 attestation from GitHub API (slsa.dev/provenance/v1)
✓ Verification succeeded!What I build
ISO 20022, end to end: pain.001, pacs.008, camt.053, and acmt.001. Adopt one library at a time.
Explore the ISO 20022 librariesPost-quantum primitives: ML-KEM (CRYSTALS-Kyber, NIST FIPS 203) and hashing, ready to pilot today.
See the post-quantum primitivesAudited Rust libraries and a code security scanner, signed and shipped with a software bill of materials.
View the toolingProof
Most downloads come from transitive dependencies in CI/CD pipelines and production services across PyPI, crates.io, and npm · Updated .
Recognition

EPAA · September 2025
Regulators are treating harvest-now-decrypt-later as a credible present risk. This paper, produced for the Emerging Payments Association Asia, sets out the structural threat and the case for coordinated action across wholesale payment rails.
Latest
TL;DR Why the FCA regulates AI through its existing principles-based, outcomes-focused rules rather than an AI rulebook, and where governance may need to adapt.
TL;DR Why the seeds behind NIST's P-256 and related curves came from lrand48, not hashes, how Claude found out in under half an hour, and why it is not a break.
TL;DR Why agentic AI breaks task-exposure forecasts that assume jobs survive automation, and how the Agentic Task Exposure score measures workflow coverage.
Tell me what you are building. The typical first response is within 48 hours.