Sebastien Rousseau

AI

AI

Applied AI in financial services — model deployment, governance, AI agents, LLM tooling, and the AI infrastructure that runs in banking environments.

67 articles

APPLIED AI

The Wholesale Payments Index in 2026: ISO 20022, Tokenised Deposits, Real-Time Rails, and Cross-Border Settlement

An index framework for measuring wholesale-payments readiness in 2026: ISO 20022 structured-address compliance ahead of SWIFT's November 2026 milestone, tokenised-deposit settlement, BIS Project Agorá cross-border atomicity, real-time rail orchestration, and liquidity efficiency. Four percentages — structured-data completeness, rail-routing optimality, settlement-finality lag, and Agorá-corridor coverage — turn payment-operations posture into supervisory-ready evidence.

APPLIED AI

The Cloud Native Banking Index in 2026: DORA, Platform Engineering, Sovereign Cloud, and Operational Resilience

An engineering blueprint for cloud-native banking in DORA audit phase. Five platform-engineering primitives — Kubernetes paved roads, Backstage portal, GitOps via ArgoCD, Open Policy Agent admission, OpenTelemetry end-to-end — produce Article 8 register evidence at the speed of the pipeline. Tested exit-execution annually for CTPP-dependent CIFs against BIA-derived RTO targets. Sovereign-cloud options (AWS European Sovereign Cloud, Microsoft EU Data Boundary, Bleu, Thales / S3NS) addressed as engineering decisions, not branding.

APPLIED AI

The Quantum-Safe Banking Index in 2026: Post-Quantum Cryptography, QKD, Crypto-Agility, and Harvest-Now-Decrypt-Later Risk

An index framework for measuring quantum-safe banking readiness in 2026: cryptographic bill of materials, hybrid TLS deployment, NIST FIPS 203 / 204 / 205 migration progress, crypto-agility primitives, and harvest-now-decrypt-later exposure across long-lived confidential data. The Board-Level Quantum Scorecard defines four exact percentages — inventory completeness, HNDL exposure, NIST migration progress, crypto-agility readiness — that turn project statuses into supervisory-ready evidence.

APPLIED AI

The Agentic AI Index for Banks in 2026: Measuring Autonomy, Governance, Auditability, and Business Impact

An engineering blueprint for agentic AI in tier-1 banks: classify by permissions not intelligence, treat every production agent as an SR 11-7 / SS1/23 model from day one, and build the five-component control plane — OAuth-scoped service accounts, deterministic semantic routing, OPA policy gates, immutable WORM audit logs, and a tested kill switch — that turns autonomous workflows into auditable evidence.

APPLIED AI

Post-Quantum Payments Infrastructure: Why Banks May Replace Rather Than Retrofit Legacy Rails

Harvest-now-decrypt-later turns today's TLS-protected payment messages into tomorrow's decrypted exposure. ML-KEM and ML-DSA are an order of magnitude larger than the RSA and ECC keys legacy rails were sized for — retrofit triggers fragmentation, latency, and HSM exhaustion. The 2026 architectural decision is whether to patch or replace before the regulatory clock runs out.

APPLIED AI

Wholesale Digital Markets in the UK: Tokenised Gilts, Settlement, and the New Champion Role

The UK Wholesale Digital Markets Champion role makes tokenised gilts, DLT-based settlement, and digital wholesale finance a national competitiveness project. The 2026 strategic question for banks is design discipline — which tokenised assets sit on which platform, how settlement risk is evidenced, and how to defend the City's wholesale role against Singapore, Switzerland, and Frankfurt.

APPLIED AI

ISO 20022 After Migration: Turning Payment Data into Banking Products in 2026

ISO 20022 after migration is not a project closeout — it's the start of a data-product opportunity. Structured addresses, purpose codes, invoice details, and rich status events become reconciliation, fraud, liquidity, and compliance products. The 2026 question is design discipline: which payment data carries which client workflow, priced for which corporate.

APPLIED AI

Digital Assets in 2026: The Infrastructure Transition Banks Cannot Delay

Digital assets are no longer an innovation bet. They are a settlement, custody, and balance-sheet architecture question. Crypto is a ~$3 trillion asset class, stablecoins have reached ~$300 billion, and tokenised real-world assets grew ~300% in 2025. The strategic task for banks in 2026 is not whether to engage but how to design infrastructure, controls, and governance for a world where money, assets, and settlement become programmable.

APPLIED AI

DORA, the EU AI Act, and Data Sovereignty: The 2026 Compliance Stack for Banks

DORA, the EU AI Act, GDPR, cloud concentration risk, and data sovereignty converge into one 2026 compliance stack for banks. The strategic question is design discipline — running compliance, AI governance, ICT-resilience evidence, and sovereign-data choices as one architecture, not five parallel programmes.

APPLIED AI

Stablecoins vs Tokenised Deposits in 2026: What Banks Actually Need to Defend

Stablecoins and tokenised deposits converge around programmable settlement but diverge on balance-sheet, legal, liquidity, and trust models. The 2026 strategic question for banks is design discipline — which money model carries which client workflow — and how to defend deposit franchise, settlement role, and regulated brand against private-issuer competition.

APPLIED AI

Programmable Liquidity in 2026: AI, Tokenised Deposits, and Real-Time Treasury Orchestration

Programmable liquidity in 2026 connects AI forecasting, tokenised deposits, real-time rails, virtual accounts, stablecoins, and TMS into one orchestration layer for working capital. The strategic question for banks is design discipline: which data, rails, controls, liabilities, and client workflows belong together — and how to evidence the orchestration to clients and supervisors.

APPLIED AI

Agentic Payments in Banking: Consent, Liability, and the New Payment UX in 2026

Agentic payments crossed from concept to live transaction in 2026 — Mastercard and Rabobank executed a Netherlands AI-agent payment via Agent Pay, AP2-style cryptographic mandates are emerging, and HM Treasury says payment-services regulation must adapt. The banking architecture question is consent, identity, liability, and pre-settlement evidence.

APPLIED AI

Quantum-Safe Payments: Why the Industry Must Act Now

Modern payment infrastructure relies on public-key cryptography. RSA, ECC, and Diffie-Hellman. To authenticate transactions, protect cardholder data, and secure messaging between financial…

APPLIED AI

Quantum Lattice Crypto: Bug in Chen's LWE Attack

Following my recent article on the [Challenges in Quantum Algorithms for Lattice-Based Cryptography][00], I am compelled to provide an update on the latest developments in [Yilei Chen's research…

APPLIED AI

Advancing AI with Multimodal LLMs: Insights from MM1

The integration of natural language processing and image recognition has resulted in the development of Multimodal Large Language Models (MLLMs). In their paper, Apple introduces the MM1, a…

APPLIED AI

RustLogs (RLG): Structured Logging Library for Rust

In the world of software development, logging plays a crucial role in understanding the behaviour of an application, diagnosing issues, and ensuring smooth operation. Rust, a systems programming…

APPLIED AI

Le Chat by Mistral AI: Multilingual Conversational AI

Conversational AI is evolving rapidly, and Mistral AI's Le Chat is leading the way in this innovation. This platform improves user experiences and challenges top AI solutions like ChatGPT. It's…

APPLIED AI

Unlocking Gemini 1.5: Google's AI Revolution Explained

Gemini 1.5 is built on a new version of the Mixture of Experts (MoE) architecture, enhancing its ability to selectively activate the most relevant pathways in its neural network. This advancement…

APPLIED AI

Àkàndé: GPT-Powered Voice Assistant for Executives

Àkàndé is an open-source Python voice assistant that chains OpenAI Whisper speech recognition, GPT-4 chat completions, and a local SQLite response cache into a voice-driven workflow — generating PDF summaries from conversation history via fpdf2 and keeping all stored data local with no cloud persistence.

APPLIED AI

Generative AI in Advertising: Amazon Rufus & Meta

Artificial Intelligence (AI) is drastically transforming the advertising landscape, offering unprecedented efficiency and effectiveness in ad campaigns. Amazon's introduction of [Rufus ⧉][00], a…

APPLIED AI

Audio Analyser: Azure Speech, NLP, and Translation Pipeline

Audio Analyser is an open-source Python tool that chains Azure Cognitive Services speech-to-text, Text Analytics NLP, and Translator into a single CherryPy web application — turning audio recordings into structured transcripts, sentiment scores, and multilingual translations without cloud lock-in beyond the Azure SDK.

APPLIED AI

AI Prompt Engineering 2024: Techniques That Work

Prompt engineering structures LLM input at inference time — no weight updates required. This article covers the techniques that proved reliable in 2024: zero-shot task framing (Brown et al., 2020), chain-of-thought reasoning (Wei et al., 2022), self-consistency sampling (Wang et al., 2022), ReAct agent loops (Yao et al., 2022), indirect prompt injection risk (Greshake et al., 2023), and applied RAG patterns from financial services.

APPLIED AI

Alien Studio: My Tech-to Art Journey in Photography

2023 was a year of profound transformation and realisation. Motivated by curiosity and a drive to innovate, I delved into cutting-edge technology, including [Rust ⧉][06], [AI ⧉][07], and [Quantum…

APPLIED AI

KyberLib: Rust CRYSTALS-Kyber for Post-Quantum

KyberLib is a Rust-based library that protects your data from the potential threat of quantum computing. Built upon the CRYSTALS-Kyber algorithm, KyberLib delivers exceptional security,…

APPLIED AI

Generative AI in 2023: How It Works, Where It Lands

Generative AI crossed from research curiosity to production deployment in 2023. GPT-4, Claude 2, Llama 2, and Mistral demonstrated that large language models could handle legal document review, code generation, and customer dialogue at human-comparable quality — raising immediate governance questions about hallucination, data leakage, and regulatory compliance in financial services.

APPLIED AI

LibMake: Rust Library Scaffold Generator

LibMake is an open-source Rust CLI tool that generates a complete library scaffold from a single command or a TOML/YAML config file — producing Cargo.toml, a doc-commented src/lib.rs, integration test and benchmark harnesses, and a GitHub Actions CI workflow covering stable/beta/nightly Rust, clippy, rustfmt, and cargo-audit.

APPLIED AI

Static Site Generator: Fastest Rust-Based SSG

In the ever-evolving digital landscape, speed and quality are of paramount importance. I was determined to find a way to quickly generate static websites that showcase high quality, accessibility,…

APPLIED AI

Automating ISO 20022 Payment Files Creation with pain001

pain001 is an open-source Python library that generates ISO 20022 pain.001.001.09 (CustomerCreditTransferInitiation) XML from CSV or SQLite input, validates output against the XSD schema, and computes CtrlSum and NbOfTxs automatically — supporting both SEPA SCT and CBPR+ cross-border message formats.

APPLIED AI

The making of the Express Transaction Credits Platform

A hands-on account of building the EXTC platform on Ethereum ERC-223 in early 2018: how the token contract was structured, what the multi-signature and time-lock primitives were designed to do, and what the experiment revealed about blockchain's practical limits for payment networks.

APPLIED AI

Unveiling a new Cryptocurrency and Faster Payment Solution

The Express Transaction Credits (EXTC) platform was an early-2018 experiment in programmable payments on Ethereum, using ERC-223 smart contracts to enable instant cross-border transfers and atomic payment flows — ideas that decentralised finance later scaled.

APPLIED AI

Understanding the Technology behind Blockchain

Blockchain is an append-only ledger secured by cryptographic hash chains and distributed consensus — a design that makes tampering computationally expensive and auditable by anyone. Ethereum extended this foundation with a programmable execution layer, turning a simple payments record into a platform for smart contracts, tokens, and decentralised finance.

APPLIED AI

Bitcoin: A Year In Review of the First Cryptocurrency

Bitcoin, often hailed as digital gold, initiated a financial revolution by introducing decentralized electronic cash and blockchain technology. It enabled online payments between individuals…