Jiaxi He — Passion. Professional. Trust.

Intro

I am an engineering manager in Singapore, currently Vice President, Group Operations at a large bank in Singapore.

I lead a team of 7–8 engineers and data scientists in operational data, developing GenAI and agentic AI solutions for banking customers and internal users — including the transformation from GenAI solutions to agentic AI solutions — with a strong focus on evaluation, governance, and operational readiness in a regulated environment.

My focus is deployment engineering: architecture, evaluation, reliability, and adoption, rather than model research alone.

Showcases

Current work, described in general terms, alongside earlier showcases from my research and teaching years — starting with the most relevant: the GenAI to agentic AI transformation.

Swipe, scroll sideways, or use the arrows to shift through the showcases.

1 / 6

Illustration of a chat bubble flowing into a connected network of agents and tools — data, a gear, a lightning action, and a target

GenAI to agentic AI transformation

Developing GenAI and agentic AI solutions for banking customers and internal users, moving from GenAI solutions to agentic AI solutions in production, regulated workflows. Instead of one assistant that only answers, an orchestrator reads the user's query, works out the goal, and routes the work to specialised sub-agents that use data and tools to complete the task.

Orchestrator pattern Sub-agents Tools & APIs Production delivery

Illustration of a phone showing a chat conversation with a voice waveform and microphone

Conversational AI for banking

Conversational AI solutions for business banking, commercial banking, and consumer banking voice use cases, with real-time guardrails for customer banking.

Chat Voice Real-time guardrails

Illustration of a call waveform being turned into a written summary document

Service support, summarised in-system

Call transcripts are summarised in-system so client service officers can log and review quickly, instead of writing summaries manually.

Summarisation Service operations Officer workflow

Illustration of an evaluation dashboard with charts, checks, and a shield

Evaluation, guardrails, and post-chat monitoring

Leading evaluation and regulatory work, including post-chat monitoring, covering summarisation quality, accuracy, safety, and production readiness — and using those results to decide what ships.

Evaluation Guardrails Post-chat monitoring Regulatory readiness

Illustration of a gallery visitor chatting with a dialogue system about framed artworks on a gallery wall

Artlogue — a dialogue system for the National Gallery

An earlier showcase from my time at the National University of Singapore: a dialogue system for the National Gallery, where visitors can chat with the system and ask questions about the artworks — an applied conversation system, built for a public cultural setting rather than a lab demo.

Earlier work Dialogue system National Gallery Conversational AI

Illustration of documents flowing into a brain-like text-analysis engine, producing charts, keyword tags, and highlighted results

Cerebrum — a text-analysis engine

An earlier showcase: a text-analysis engine for making sense of written text at scale, turning documents into structured, searchable insight instead of leaving them as unstructured prose.

Earlier work Text analysis Natural language processing

How the agentic system works

The pattern behind the first showcase, step by step. Select a stage to read what it does.

From a user's query to one combined answer

Each stage has its own job — and its own prompt that tells it how to do that job.

Orchestrator

Understands the user's input, breaks it into the tasks needed, and works out which sub-agents are required to generate the answer. Simple questions that need no knowledge-base information are answered directly here.

Prompt: the orchestrator has its own prompt, describing each sub-agent — what it is for, when to use it, and what it returns — so it can plan the work.

The sub-agents

Each sub-agent covers one broad area of banking help, so questions are handled in general terms. Select one to see what it handles.

Account Information. General help with account-related questions — for example, questions about accounts and the information or services connected to them.

Tutorials

Earlier teaching work, kept here — my open-sourced tutorial and the classes I taught, with the original videos on YouTube.

An Introductory Course on Deep Learning with Keras

Part 1: Deep Learning Foundations

This video talks about the foundations of deep learning. When you finish this video, you will understand the major components of Deep Neural Networks, like perceptron, activation functions, forward and backpropagation, as well as the optimizers.

Watch on YouTube

Part 2: Deep Learning with Keras

In this 1-hour video, you will be able to build and train a fully connected deep neural network in Keras for the classic handwritten-digit problem. You will understand the key parameters in a neural network's architecture.

Watch on YouTube

Dialogue systems research

Chit-chat, task-oriented, and question-answering bots — including retrieval-based conversation, semantic matching, and multi-turn conversation.

Experience

Sep 2023 – presentVice President, Group Operations — DBS, Singapore
Aug 2021 – Sep 2023Senior Consultant — Ernst & Young, Singapore
Dec 2018 – Jul 2021Senior Data Scientist — National University of Singapore
Dec 2017 – Dec 2018Consultant — Deloitte, Shanghai, China
Apr 2017 – Nov 2017Data Scientist Intern — Immersive Pty Ltd, Melbourne, Australia

Education

2012 – 2017Ph.D. in Computer Science — Swinburne University of Technology, Australia
2010 – 2012M.Sc. — University of Southern Denmark, Denmark
2006 – 2010B.Sc. — Northwestern Polytechnical University, China