Hello, I'm

Xiao Jun Ang

Agentic AI

Python

LLMs

LangChain

Docker

RAG

About Me

I'm a Software Development Engineer II at AMD, focused on end-to-end Agentic AI development — from prompt and context engineering to production deployment. I design and build AI agents (multi-step workflows, tool use, function calling), develop RAG pipelines and context-management strategies, and deploy them into live enterprise workflows as a forward-deployed engineer, working directly with stakeholders to refine use cases and ship reliable systems.

I recently completed AI Singapore's national AI Apprenticeship Programme (AIAP) and graduated with a Master of IT in Business (AI) from Singapore Management University. Outside of work, I build broadly across classical ML, computer vision, and MLOps through personal projects. My background in Pharmaceutical Science (NUS, Honours Distinction) gives me a unique lens — bridging rigorous scientific methodology with modern machine learning to drive data-informed decisions and scalable AI solutions.

4+ Years in Science & Data

10+ ML Tools & Frameworks

3 Academic Qualifications

Featured Projects

Computer Vision E-Commerce

Standardising e-commerce product images with AI-powered scoring, segmentation and background inpainting

GitHub

VisionCommerce

AI pipeline that tackles visual fragmentation for e-commerce platforms — automatically scoring, segmenting, and standardising product images at scale.

✦ Fine-tuned DINOv2 evaluates product image quality based on clarity, background, and aesthetics
✦ SAM3-based segmentation generates precise product masks for targeted background editing
✦ FLUX.1-FILL diffusion model inpaints consistent, high-quality backgrounds across the catalogue

PyTorch DINOv2 SAM3 FLUX.1-FILL Streamlit Computer Vision

MLOps Time Series

Predicting US flight arrival delays with end-to-end MLOps — from data ingestion to real-time monitoring

GitHub Live Demo

Flight Delay Prediction

End-to-end MLOps system predicting US flight arrival delays across 3 aviation datasets, with a live Streamlit dashboard for real-time exploration.

✦ Medallion architecture (Bronze/Silver/Gold) with PySpark for scalable data preprocessing
✦ Airflow-orchestrated pipeline with MLflow experiment tracking and email alerting
✦ Evidently AI drift monitoring with monthly evaluation cycles on production data

PySpark MLflow Airflow Docker Evidently AI Streamlit

NLP Regression Classical ML

Benchmarking job salaries with 43% better accuracy than statistical baselines using LLMs and neural networks

GitHub Live Demo

Job Salary Prediction

ML system for salary benchmarking trained on 34K US LinkedIn job postings, achieving R² 0.68 with an MLP — outperforming all classical baselines.

✦ Gemini 2.0 Flash extracts structured requirements (skills, education, certifications) from raw job descriptions
✦ SBERT embeddings (all-MiniLM-L6-v2) encode semantic job features for downstream modelling
✦ MLP with Bayesian hyperparameter tuning achieves 43% MAE reduction over statistical baselines

PyTorch XGBoost SBERT Gemini scikit-learn NLP

Work Experience

Software Development Engineer II

AMD

Aug 2026 – Present

Building end-to-end Agentic AI — from prompt and context engineering to production deployment — and integrating AI agents with enterprise systems as a forward-deployed engineer.

Python Agentic AI LLMs RAG Function Calling Context Engineering Azure OpenAI Docker

Design and build AI agents with multi-step workflows, tool usage, and memory, implementing function calling and structured outputs.
Develop RAG pipelines and context-management strategies, optimising reliability, latency, and response quality.
Engineer system and task prompts and retrieval strategies to reduce hallucinations and improve consistency, backed by evaluation and testing frameworks.
Deploy AI solutions into live business workflows (Forward Deployment Engineering), integrating with APIs, databases, and ERP systems.
Work directly with stakeholders to refine use cases, then monitor, debug, and continuously improve production systems.

AI Engineer (AIAP)

AI Singapore

Jan 2026 – Jul 2026

Graduated from a competitive national AI programme, then deployed on a 100E industry project building a multi-agent LLM platform for a global semiconductor company.

Python Multi-Agent LLM RAG Guardrails LLM-as-Judge Red-Teaming GitLab CI/CD Docker GCP

Built and shipped production ML and LLM systems end-to-end: agent architectures, inference APIs, containerised deployment, CI/CD, and evaluation pipelines.
On a 100E industry project, built a multi-agent LLM platform that lets business teams investigate operational data in natural language and receive evidence-backed reports.
Designed trust guardrails — all figures deterministically computed and validated over validated SQL, with the system abstaining rather than hallucinating.
Built the evaluation and observability loop: ground-truth scoring, LLM-as-judge, span-level trace analysis, and red-teaming (prompt injection, misuse) before release.
Established a self-improving workflow where AI coding agents log decisions to a knowledge base, distilled into reusable engineering rules.
Partnered directly with business and IT stakeholders on requirements, reviews, and production handoff.
Foundations coursework: CNNs, transfer learning, LSTMs, transformers, VAEs, and diffusion models; deployed with GitLab CI/CD, Docker, and GCP.

Machine Learning Engineer

Beeva AI

Sep 2025 – Jan 2026

Developed and fine-tuned deep learning models in PyTorch for computer vision, with production-grade MLOps workflows supporting reproducible research.

PyTorch Computer Vision MLflow DVC Docker Deep Learning

Developed and fine-tuned deep learning models in PyTorch for computer vision applications, with emphasis on generalizable feature learning and robust validation.
Implemented production-grade MLOps workflows (MLflow, DVC, Docker) to support reproducible research and scalable deployment.
Optimized models for latency, inference cost, and deployment constraints, balancing experimental performance with real-world usability.

Research Officer

A*STAR – Nutrition & Digestive Health

Oct 2023 – Jan 2026

Conducted PBPK/PK-PD modelling and quantitative LC-MS/MS analysis to support drug bioavailability research and evidence-based decision-making.

PBPK Modelling LC-MS/MS Python R Data Analysis ADME

Designed and executed in vitro biological and biochemical models (Caco-2 and SHIME®) to quantify compound bioavailability, transport, and functional performance.
Conducted quantitative LC-MS/MS analysis to support PBPK and PK/PD-informed models assessing pharmaco/nutrico/toxicokinetic behaviour of drugs and food ingredients.
Translated biological mechanisms (ADME, transporter kinetics, enzyme interactions) into model-ready parameters to improve clinical prediction accuracy.

Product Development Lead

Craft Health

Jun 2022 – Sep 2023

Led data-driven product development for pharmaceuticals and nutraceuticals, managing a team of five across multiple client projects.

Data Analytics Market Research Product Development Team Management

Led data-driven product development for pharmaceuticals, nutraceuticals, and food & beverage — translating technical and market requirements into actionable plans.
Conducted quantitative market research and trend analysis to identify consumer needs and generate product innovation strategies.
Managed a team of five formulation associates, applying structured methodologies and quality control metrics to ensure reproducibility of outcomes.