FINANCE × AI × REAL-WORLD IMPACT
I’m Nils, an AI architect and hands-on builder. From bank-scale GenAI to mortgage review workflows, I turn complex financial problems into AI people can actually use.
Nils Liu Head of AI @ AAA Lendings
SELECTED WORK
A selection of prototypes exploring where AI meets everyday work.
From browser actions to structured reports, exploring how agents can assist identity checks with a trace for human review.
A collaborative, multilingual research workflow that gathers news, keeps sources, and makes room for follow-up questions.
Sunny & Rainy Forest: 28,000 Days explores insurance and financial planning through life simulation. Lead developer: 邱顏敏.
A Google Meet participant that helps capture conversations and organize interview feedback. Lead developer: 孫旖岑.
HOW I CAN HELP
AI leadership, platform strategy, and focused collaboration with the people building what comes next.
Work through the business problem, platform decisions, and the data and governance constraints that shape the solution.
Focused conversationsTurn the hardest assumptions into a working prototype, then use evaluations to decide what deserves further investment.
2–6 weeksHelp your team learn through its own challenges, from retrieval and agent workflows to evaluation and handoff.
0.5–2 daysFrom chemistry labs in Taiwan to research and industry in Germany, then into financial technology. I’m drawn to problems between disciplines, and technology that makes people’s work better.
Now Head of AI at AAA Lendings. Previously, I led CathayGPT, AI Coach, and overseas AI Agent platforms in banking, connecting architecture, governance, and real adoption.
3-layer Agent architecture with GraphRAG and Rubrics-based evaluation.
Sole inventor on all 6. 2 additional invention patents under review.
iThome Hello World 2024+2025, DevFest Taipei 2025.
Building AAA's AI function from the ground up. Leading a 3-person engineering squad to deliver an end-to-end mortgage AI review platform on Encompass data, covering document intake, extraction, validation, review workflow, and submission / underwriting handoff.
Led production GenAI and agentic AI platforms for financial-services users: CathayGPT, AI Coach, and the overseas AI Agent platform. Owned solution architecture, rollout, governance, and measurable adoption.
Consulted at a Swiss financial advisory firm; built the GenAI risk analytics module 'Narrator'.
Joined via Management Associate program. Led Python automation and ML enablement for 200+ staff.
Managed tech-transfer projects and pilot-line implementations, coordinating cross-border teams for automation acceptance.
Conducted advanced materials and nano-synthesis research at a Nobel Laureate lab (Holland) and Fraunhofer (Germany).
Pursued a Master's degree in Germany, gaining international academic and cross-cultural experience.
Graduated from National Taiwan University, building a rigorous scientific foundation.
Python · SQL · GCP (Cloud Run / Vertex AI) · Serverless Architecture · Agentic Workflows / MCP · RAG / GraphRAG / Agents / Evals · React / Next.js / TypeScript · LangChain / LlamaIndex · PostgreSQL / Vector DBs
Scale and compute win. I build systems that leverage foundation models rather than fight them.
One-off solutions don't survive. I design for reuse, governance, and team autonomy from day one.
If you can't measure it, you can't ship it. Rigorous evaluation is the foundation of every project I touch.
LLMs produced unstructured, unauditable investment recommendations. This patent enforces structured output that meets financial compliance requirements.
Enterprise prompts were managed ad-hoc with no versioning. This patent establishes a version-controlled prompt optimization mechanism.
Traditional keyword search fails on unstructured enterprise documents. This patent introduces a semantic retrieval architecture for accurate information access.
Different bank departments needed custom AI behaviors on the same underlying model. This patent solves the multi-scenario deployment scaling problem.
Invention Patent PendingCross-system data queries were fragmented and costly to integrate. This patent provides a unified query interface to reduce engineering overhead.
Unstructured enterprise knowledge was inaccessible to AI systems. This patent builds a reasoning-capable knowledge structure so AI genuinely understands business context.
Invention Patent PendingNOTES & PERSPECTIVES
Notes on AI, products, and what happens when technology meets real life.
LET’S TALK
An AI platform, a leadership opportunity, a workshop, or a useful conversation. Tell me what you have in mind.
wulingteen@gmail.comTAIPEI, TAIWAN · GMT+8
