About Jayden

The work has moved, but the habit stayed.

I started in full-stack development, moved into enterprise banking systems and technical management, and now work in AI engineering and data science consulting at Kyndryl. The thread is simple: understand the system, reduce the ambiguity, and ship something useful.

The homepage capability map is the index. This page is the lower level version: the projects, coursework, systems, and judgment behind those labels.

View portfolioGitHub

Capability notes

High-level map, lower-level evidence.

The capability map stays compact on the homepage. Here, each label expands into the work, academic route, and production constraints behind it.

01 / Capability

AI Systems Engineering

Production AI apps, enterprise workflows, and deployment discipline.

Turning AI capability into shipped systems.

My current work sits at the layer where prototypes become production workflows: requirements discovery, architecture trade-offs, cloud deployment, stakeholder alignment, and delivery across enterprise environments.

  • Led enterprise AI initiatives at Kyndryl from discovery through architecture, deployment, and team enablement.
  • Delivered AI systems across fintech, media, security, and legal document workflows.
  • Balanced latency, compliance, cost, and reliability constraints instead of treating models as isolated demos.

Tools and concepts

PythonFastAPIGCPVertex AIDockerKubernetes

02 / Capability

Agentic Workflows

Tool-using agents, multi-step orchestration, and practical automation.

Designing agent flows around real user tasks.

I treat agents as workflow systems, not chat wrappers. The work is deciding what should be retrieved, what should be planned, what should be delegated to tools, and where human review needs to stay in the loop.

  • Architected Google ADK multi-agent workflows for a Singapore fintech navigation and FAQ chatbot.
  • Used LangChain orchestration for retrieval and task flow decisions across RAG and GraphRAG paths.
  • Designed agent behavior around MAS compliance, data sovereignty, and measurable query resolution.

Tools and concepts

Google ADKLangChainAgentsTool UsePrompt Design

03 / Capability

Retrieval Systems

RAG, GraphRAG, vector search, and hybrid lexical-semantic retrieval.

Building retrieval around document structure, not naive chunks.

The retrieval work I care about starts with the shape of the knowledge. For legal, banking, and navigation tasks, structure matters: clause hierarchy, graph relationships, lexical precision, semantic recall, and calibrated ranking.

  • Built clause-level metadata with clause IDs, sub-clause hierarchy, and structural embeddings for legal MSA review.
  • Implemented PostgreSQL Full-Text Search with pgvector cosine similarity and Reciprocal Rank Fusion.
  • Evaluated RAG and GraphRAG trade-offs for enterprise FAQ and mobile navigation use cases.

Tools and concepts

RAGGraphRAGPostgreSQL FTSpgvectorRRF

04 / Capability

ML Systems

Model evaluation, inference pipelines, MLOps, and explainability.

Owning model behavior beyond training code.

ML systems work means choosing models, defining evaluation, running pipelines, explaining results, and making outputs usable for operators. I focus on the seams between modeling, infrastructure, and decision support.

  • Owned anomaly detection delivery across 45M+ records using DBSCAN, autoencoders, and One-Class SVM evaluation.
  • Built Cloud Composer production pipeline and PowerBI executive dashboard for security incident workflows.
  • Used LIME explainability to make anomaly outputs inspectable by client stakeholders.

Tools and concepts

TensorFlowScikit-learnCloud ComposerBigQueryLIME

05 / Study

Applied Deep Learning

Deep learning, AI coursework, and systems-level academic grounding.

Strengthening practice with formal AI study.

I am pairing enterprise AI delivery with a part-time Master of Computing at NUS and focused coursework in deep learning, machine learning systems, and artificial intelligence.

  • Taking Master of Computing, Computer Science and AI at NUS.
  • Academic route includes Stanford Online CS230 Deep Learning.
  • NUS modules include CS5462 Machine Learning Systems and IT5005 Artificial Intelligence.

Tools and concepts

CS230CS5462IT5005Deep LearningAI Systems

06 / Practice

AI Transformation

Education, adoption, stakeholder translation, and product judgment.

Helping teams adopt AI without losing judgment.

At UOB, the work was not only building tools. It was teaching engineers, translating AI capability for executives, creating champions, and turning ambiguous interest into safer, more useful adoption.

  • Led AI adoption across 200+ staff through 20+ workshops, technical sessions, and executive presentations.
  • Built an AI-powered code review assistant with Groq, Flask, Bitbucket API, and Next.js.
  • Created a centralized prompt library and mentored internal AI advocates to scale adoption.

Tools and concepts

WorkshopsPrompt EngineeringGroqFlaskNext.js

Proof point

case file

AI-Powered Legal Document Intelligence - MSA Redlining System

End-to-end AI redlining system for legal MSA review, replacing brittle rule-based parsing with LLM-driven clause ingestion across Playbook, Draft Versions, and Final State.

  • Designed clause-level metadata with clause ID, sub-clause hierarchy, and vector embeddings at structural granularity.
  • Built reusable DOCX/PDF to JSON, MD, and JSONL parser with confidence_score and importance_score per clause diff to gate retrieval.
  • Combined PostgreSQL Full-Text Search using ts_rank_cd with pgvector dense cosine similarity, fused through Reciprocal Rank Fusion.
  • Engineered modular prompt system using top-k context and diffs to generate redlined clauses with justifications and Playbook citations.
  • Captured clause-level feedback to seed future evaluation and fine-tuning datasets.
  • Deployed full pipeline end-to-end on GCP and at work.
LLMsDOCX/PDF parsingPostgreSQL FTSpgvectorRRFGCP

Academic route

Formal study alongside production work.

Master of Computing, Computer Science and AI, NUS

CS230 Deep Learning, Stanford Online

CS5462 Machine Learning Systems, NUS

IT5005 Artificial Intelligence, NUS

Journey

The path still matters.

Full-stack roots

Interfaces, APIs, product workflows, and the habit of shipping usable software.

Enterprise banking

3DS, cards systems, legacy integration, incident response, and reliability under real production constraints.

AI transformation

UOB education, internal tooling, prompt practices, executive translation, and adoption across engineering teams.

Enterprise AI systems

Kyndryl AI delivery across agents, retrieval, live captioning, anomaly detection, and client-facing architecture decisions.