Explore my website · Résumé (PDF) · LinkedIn · Email me
I'm Goh Kun Ming, a Research Engineer intern at NCS · Pragnition Labs and a Year 3 Applied AI & Analytics student at Singapore Polytechnic, based in Singapore.
I work across machine learning, computer vision, retrieval-augmented AI, and data products. I enjoy taking a research question through the whole build: preparing data, comparing models, evaluating limitations, and connecting the result to an application someone can use. My longer-term direction is AI research and academia, including teaching and mentoring.
Explore my work for research engineering, applied AI/ML, and data-product opportunities.
Selected work / More projects / Experience / Technical toolkit / How I build / Beyond the projects / Contact
-
Research & evaluation
Hypothesis-driven experiments, classical baselines, generative modelling, and bounded quantum-ML comparisons
Hybrid generative models -
Applied AI systems
Vision pipelines, model serving, APIs, and user-facing applications
EstateScope AI · VeggieAI -
Data & decisions
Data validation, exploratory analysis, statistical reasoning, and clear visual explanations
HDB resale analytics
Five projects spanning generative research, AI applications, housing analytics, property-price prediction, and time-series forecasting. Explore the linked repositories for methods and results.
Research question: how do classical and circuit-based latent priors compare in a generative model?
Compared a classical GAN with 3-, 5-, and 7-qubit hybrid variants on binary MNIST, with shared configuration and FID/KID evaluation utilities. The repository documents the comparison, reproducibility steps, and limitations.
Python · TensorFlow · Qiskit · Optuna
Modelling focus: estimate property prices from complementary input types.
Built a Flask prototype for US property-price estimation using structured attributes and optional listing descriptions and images. The application exposes per-modality estimates, combines available inputs, and supports authenticated prediction history and housing analytics. The repository documents held-out evaluation of the NLP component.
Python · Flask · scikit-learn · TensorFlow · SQLite · Docker
Explore the prediction application →
Product focus: connect image classification to a usable, testable application.
Built a Flask application with separate model serving, authentication, prediction history, a model registry, and Docker tooling. CI includes backend and model-service tests, registry validation, and local stack smoke checks.
Flask · TensorFlow Serving · SQLite · Docker · GitHub Actions
Decision focus: make Singapore housing data easier to explore and explain.
Prepared a cleaned dataset of 202,461 resale records and Tableau views covering prices, lease tenure, flat types, and location. The repository preserves cleaning, geospatial enrichment, and validation steps so the analysis can be inspected and reproduced.
Python · pandas · Tableau · Geospatial analysis
Forecasting focus: turn historical consumption into a reproducible modelling workflow.
Developed a time-series workflow for monthly gas, electricity, and water consumption, combining data validation, outlier handling, ARIMA/SARIMAX modelling, residual analysis, and a configured 60-month forecast horizon. Each target is modelled separately, with evaluation and limitations documented alongside the code.
Python · pandas · statsmodels · scikit-learn · Time-series analysis
Explore the forecasting workflow →
More of my work across decision support, language algorithms, data structures, full-stack development, and reinforcement learning.
Built a Dash dashboard for GPA, attendance, cohort trends, and survey analysis. Configurable rules mark a tier as high when both GPA and attendance fall below their thresholds, medium when one does, and low when neither does—making the decision logic easy to inspect.
Python · Dash · Plotly · pandas
Created a Python CLI combining syllable-aware Markov generation, thesaurus-based text transformations, and word-level WAV narration. It explores how language can be generated and reshaped with explicit algorithms and a 5–7–5 target.
Python · Markov chains · CLI · WAV audio
Built a text-recovery CLI using prefix tries, wildcard matching, and edit-distance search to suggest restorations for damaged newspaper text. Ranked candidates and graph views make the search process inspectable.
Python · Prefix tries · NetworkX · Matplotlib
Developed a gamified fitness web application with authenticated Express APIs, user-scoped MySQL data, challenges, and game progression. Browser-flow tests cover key journeys through the application.
JavaScript · Node.js / Express · MySQL · JWT · Playwright
Developed a reinforcement-learning lab for comparing Standard, Double, Dueling, and Rainbow-style DQN on a simulated pendulum. Continuous torque is mapped into discrete actions, with reusable environment wrappers, replay buffers, training, and evaluation components. Seeded configurations and documented limitations make the experiments easier to inspect and reproduce.
Python · TensorFlow / Keras · Gymnasium · NumPy · Reinforcement learning
Browse all public repositories →
Research Engineer · NCS · Pragnition Labs
Internship · Oct 2026–present · Singapore
Turn business and technical requirements into research questions, testable hypotheses, and AI prototypes. Evaluate performance, reliability, latency, cost, security, and maintainability; communicate findings and build reusable components.
AI Research Intern · RSAF RAiD · AETHER
Apr–Aug 2026
Researched multimodal perception, geospatial data, neuromorphic vision, RF sensing, and language/vision-language reasoning. Built reproducible evaluations and documented methods, findings, and limitations for research handover.
AI & Data Solutions Researcher · Singapore Polytechnic · Department of Communications
Part-time · Mar 2026
Built role-aware departmental workflows, a Supabase data architecture, decision-support tools, and AI assistance, supported by release, access-control, and usability checks.
Co-Founder, Data & Intelligent Systems · Wagglo
Feb–Mar 2026
Designed compatibility scoring and recommendation workflows for a pet-owner matching concept, alongside privacy-conscious survey analysis to inform product priorities.
Student Developer · Singapore Polytechnic · School of Computing
Sep 2025–Mar 2026
Developed an institutional Open House assistant using Amazon Bedrock, hybrid retrieval, reranking, persistent conversation state, and controlled user evaluation, with a FastAPI backend and Next.js interface.
Full journey on my website → · LinkedIn experience →
-
Machine learning & forecasting
scikit-learn · XGBoost · statsmodels · ensemble learning · ARIMA / SARIMAX · Optuna -
Deep learning & generative models
PyTorch · TensorFlow / Keras · CNNs / ResNet · SimpleRNN / LSTM / GRU · GANs / VAEs -
Reinforcement learning
Standard / Double / Dueling / Rainbow DQN · experience replay · target networks · seeded evaluation -
Vision & multimodal research
OpenCV · Ultralytics YOLO · ONNX · multimodal perception · geospatial data · RF sensing -
Simulation
Unreal Engine 5 · AirSim · RGB / thermal / LiDAR experiments · sensor-fusion comparisons -
Quantum & generative research
Qiskit / Aer · latent-prior experiments · FID / KID evaluation · Jupyter -
Statistics & optimisation
Gradient descent · PCA · hypothesis testing · ANOVA / Tukey HSD · bootstrap analysis · cross-validation -
Clustering & evaluation
K-Means · agglomerative clustering · DBSCAN / OPTICS · error analysis · leakage checks · distribution drift -
Retrieval & language
Amazon Bedrock · OpenAI / Cohere embeddings · pgvector · BM25 · reciprocal-rank fusion · query rewriting · reranking · hallucination grading -
Data engineering & analysis
pandas · NumPy · PySpark · ETL · star-schema warehouses · data validation · geospatial enrichment -
Visualisation & dashboards
Matplotlib · Seaborn · Plotly / Dash · Tableau · Power BI · NetworkX -
Algorithms & text processing
Prefix tries · wildcard matching · edit-distance search · Markov chains · tokenisation
-
Languages
Python · SQL · JavaScript · Rust · HTML / CSS -
Application engineering
FastAPI · Flask · Node.js / Express · React / Next.js · REST APIs · Zod -
Databases
PostgreSQL · MySQL · SQLite · Supabase -
Cloud & model serving
AWS / EC2 · Linux / Ubuntu · Docker · TensorFlow Serving · model registries -
Access & data governance
JWT · role-based access · row-level security · consent records · data-lifecycle design · audit logging -
Testing & delivery
pytest · Playwright · Jest · Supertest · Ruff · Git · GitHub Actions · CI/CD · technical documentation
My professional certificate programmes and further coursework extend my toolkit in these areas:
-
AI agents & orchestration
LangChain · LangGraph · CrewAI · AutoGen · BeeAI · tool calling · agent memory -
Retrieval & agent interfaces
LlamaIndex · FAISS · ChromaDB · Gradio · Model Context Protocol (MCP) · FastMCP -
Transformers & LLM adaptation
Hugging Face Transformers · PEFT · LoRA / QLoRA · quantisation · instruction tuning -
Alignment & foundations
RLHF · PPO · DPO · reward modelling · linear algebra · multivariate calculus · AI infrastructure & operations
Skills & experience on LinkedIn → · Explore my credential record →
Stakeholder alignment → data → baseline & evaluation → product integration & feedback → validation & handover → monitoring & iteration.
I start by engaging stakeholders to understand the users, decisions, success criteria, and practical constraints. I translate those needs into requirements, inspect data quality and leakage, and compare a baseline before connecting the model to an API and interface.
Stakeholder engagement continues throughout the build: review prototypes together, explain trade-offs, gather feedback, and agree on acceptance criteria. Before release, I check behaviour, usability, access controls, failure cases, latency, and cost. Clear documentation and handover support adoption; feedback and monitoring guide the next iteration.
Literature & gap → question & hypothesis → study design → baselines & experiments → analysis & limitations → reproducibility & communication.
I begin with what is already known and what remains uncertain. I define a testable question, choose comparison conditions and evaluation criteria, and keep data splits and experiment settings explicit. I use ablations and error analysis to examine the result, report uncertainty and negative findings, and document enough detail for others to inspect or reproduce the work. The findings may refine the hypothesis or motivate the next study.
Diploma in Applied AI & Analytics · Singapore Polytechnic
2024–2027 · In progress, Year 3
My studies span machine learning, data engineering, deep learning, cloud computing, and practical AI projects.
Singapore-Cambridge GCE O-Level · Yuhua Secondary School
2020–2023
My professional certificates span AI engineering, agentic systems, deep learning, advanced analytics, and practical AI use.
-
IBM AI Engineering Professional Certificate
Machine learning, deep learning, and generative AI engineering -
IBM RAG and Agentic AI Professional Certificate
Retrieval-augmented generation, tool use, MCP, and multi-agent workflows -
IBM Deep Learning with PyTorch, Keras and TensorFlow Professional Certificate
Neural networks, transfer learning, model training, and evaluation -
Google Advanced Data Analytics Professional Certificate
Statistical inference, regression, machine learning, and data storytelling -
Google AI Professional Certificate
Responsible AI use for research, communication, data analysis, and application building
Additional specialisation: Mathematics for Machine Learning — Imperial College London, issued October 2026; linear algebra, multivariate calculus, and PCA.
View credentials on LinkedIn →
My interests include intelligent systems, AI safety and governance, quantum computing, and complexity science. Education, practice, and service guide how I want to contribute: keep learning, test ideas carefully, and make knowledge useful to others.
Outside research and code, I enjoy gaming, music, discovering places to eat, coffee, time with family and friends, and travel. The Minecraft-inspired visuals are a nod to the server my polytechnic friends and I ran together in Year 2. I like building things that invite exploration.
If your team is working on applied AI, research engineering, or data products, I'd be glad to discuss how my experience could contribute.
Email Kun Ming · Connect on LinkedIn · Read my résumé · Explore my website
Curious by default. Clear by design.