Ck Wee
Data Scientist
26+ years exp
7+ years ML exp
Assembly
SQL
Python
Data Pipeline & Feature Engineering: 5/10
Model Architecture & Training: 4/10
Experimentation & Evaluation: 4/10
Active 3 days ago
+61 (048) 0151328
Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
LLM-focused ML engineer (mid-level) specializing in building applied RAG and tabular risk-scoring systems with integrated APIs and simple model-serving artifacts. The strongest proven skill is end-to-end applied system integration - demonstrated by the NL2SQL async RAG pipeline using OllamaEmbeddings, Chroma retrieval, sqlglot safety checks and FastAPI endpoints. The public code shows clear applied engineering but lacks advanced research work, distributed training, quantization/GPU optimization or formal experiment tracking infrastructure.
Technical skills
Assembly
SQL• Senior • 26y+
Python• Senior • 5y+
Python
SQLAlchemy
Pydantic
Uvicorn
Beautiful Soup
FastAPI
Databases
Azure SQL Database
SAP HANA
Chroma
Oracle• 26y+
MS SQL• 17y+
Databricks
Snowflake
AI/ML
AutoGen
LangGraph
Ray
Replicate
Spark
LangChain
NumPy
Pandas
Streamlit
XGBoost
Scikit-learn
Anomaly Detection• 7y+
NLP• 7y+
Reinforcement Learning• 5y+
Time Series Forecasting• 3y+
Computer Vision
dbt
DevOps
AWS
Red Hat
VMWare
Rest API
Vector
Azure
Analytics
Power BI
Robotics
Digital Twin• 3y+
Timeline
Asset Health Developer (CM Data Scientist)
•
Middle
BHP
•
Contractor
Built predictive maintenance and condition monitoring models for mining and plant equipment using Python and time-series approaches. Developed asset health models using Snowflake and applied anomaly detection and failure analysis to improve reliability and reduce downtime. Produced proof-of-concept computer vision to detect driver drowsiness from facial data and designed an anomaly-detection orchestration workflow and agentic AI architecture for autonomous fault detection and decision support.
Python
Time Series Forecasting
Snowflake
Anomaly Detection
Computer Vision
Senior Data Scientist
•
Senior
Stanmore Resourcing
•
Full-Time
Served as the sole AI specialist and led AI strategy work across departments by running feasibility workshops to identify high-impact use cases. Created governance documentation and SOPs for data quality, pipelines, access control, and model deployment across Azure Synapse, Microsoft Fabric, and MSSQL-based environments. Delivered NLP and time-series proof-of-concepts, including vector similarity analysis and anomaly-focused trend monitoring for business reporting.
Python
Databricks
NLP
dbt
MS SQL
Time Series Forecasting
Anomaly Detection
Vector
Senior Data Scientist
•
Senior
Urban Utilities
•
Full-Time
Developed an operational resiliency system using machine learning to support intervention planning, prevention, and service recovery. Implemented monitoring and anomaly detection over thousands of water and sewer assets using time-series data such as flows, pressures, overflow signals, and usage patterns. Built a prediction system for future pipe choking events and developed a Python-based digital twin for network performance monitoring and anomaly detection.
Python
Time Series Forecastingsince 2023
Anomaly Detection
Digital Twin
Senior Database administrator
•
Senior
RACQ
•
Contractor
Provided contract DBA support primarily for Oracle databases and secondary support for MSSQL systems for BAU operations and application support. Worked with team leads to formulate and execute procedures to migrate in-house RAC 19c environments to Oracle Cloud Infrastructure. Managed migration progress for UAT RAC databases and supported successful handover to team leadership.
Oracle
MS SQL
SQL
University of Southern Queensland
Doctoral Degree (PhD) •
Artificial Intelligence
Data Scientist
•
Middle
Caterpillar
•
Contractor
Supported development of AI models for autonomous mining hauling vehicles for an electrification machinery initiative. Continued R&D for reinforcement-learning-based control approaches in Python and worked on scheduling optimization using linear/constraint programming. Conducted research using simulated traffic movement with QGIS and explored optimization and reinforcement learning integration with discrete event simulation tools to support mining assignment planning.
Pythonsince 2021
Reinforcement Learning
Linear
QGIS
Data Engineer
•
Middle
eHealth Queensland
•
Full-Time
Established and supported healthcare data replication into an Azure Synapse-based warehouse using Azure ingestion services and supporting analytics publishing. Researched and developed a GP referral triaging approach using medical text analytics and machine learning techniques, including outlier detection and association-rule mining. Wrote SQL views for downstream project requirements and assisted with Power BI performance troubleshooting and report deployment.
MS SQL
SQL
NLPsince 2019
Anomaly Detectionsince 2019
Senior DBA/Data Engineer
•
Senior
SPARQ Solutions / Energy Queensland Limited
•
Full-Time
Provided long-term database and data engineering support in a BI-focused environment, including administration support for large fleets of Oracle and Microsoft SQL Server databases. Managed Oracle DBA responsibilities such as backup/recovery, cloning, patching, and installation of database modules, and supported Oracle replication technologies. Delivered SQL Server administration tasks including mirroring, log shipping, and backup configuration, and led initiatives to explore machine learning for utility load forecasting and predictive maintenance using R and Azure analytics services.
Oracle
MS SQLsince 2009
SQL
Griffith University
Non-Degree Program •
Australia Migration Law
Database/System Administrator
•
Middle
Department of Justice
•
Contractor
Provided contract Oracle DBA administration and application support for JAG databases, handling performance checks, troubleshooting, and database cloning. Set up Oracle Streams for change data capture across multiple environments and supported Oracle BPEL and SOA installation and configuration on Linux servers. Managed RAC installation activities on Linux VMWare for application-team testing and supported ongoing application operations for critical systems.
Oracle
SQL
Linux System/ Database Administrator
•
Middle
SPARQ Solution
•
Contractor
Performed database administration on Oracle 10gR2 across non-production and production environments and supported Oracle options such as Data Guard. Conducted disaster recovery work using RMAN and Data Pump and supported EM Grid Control and Oracle agent/service configuration. Completed Linux system administration and UNIX scripting tasks to support DBA operations and troubleshoot production shell scripts tied to data management.
Oraclesince 2007
SQLsince 2007
Curtin University of Technology
Master's Degree •
Information Technology
University of London
Bachelor's Degree •
Computing and Information Systems
Singapore Polytechnic
Other •
Civil Engineering (Structural engineering and design)
Middle AI/ML Engineer
Confidence: High LLM Engineer
LLM-focused ML engineer (mid-level) specializing in building applied RAG and tabular risk-scoring systems with integrated APIs and simple model-serving artifacts. The strongest proven skill is end-to-end applied system integration - demonstrated by the NL2SQL async RAG pipeline using OllamaEmbeddings, Chroma retrieval, sqlglot safety checks and FastAPI endpoints. The public code shows clear applied engineering but lacks advanced research work, distributed training, quantization/GPU optimization or formal experiment tracking infrastructure.
Model Architecture & Training
4/10
How well models are designed and trained
Reasonable applied model-training pipelines for tabular insurance tasks using XGBoost with held-out evaluation and persisted artifacts; no custom neural architectures or advanced training infrastructure.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/train_model.py: XGBClassifier training, eval metrics, model and metrics persisted to data/
AI-Risk-Scoring-for-Commercial-Insurance/marine_cargo/train_model.py: similar XGBClassifier training and evaluation pipeline
Data Pipeline & Feature Engineering
5/10
How data is prepared for models
Well-structured synthetic data generation and batch feature engineering that mirror production feature-rollup jobs; clear attention to realistic signal injection into synthetic datasets.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/generate_synthetic_data.py: realistic telematics event simulation and claims generation
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/feature_engineering.py: rolling-window aggregation and label join for driver-level features
AI-Risk-Scoring-for-Commercial-Insurance/marine_cargo/generate_synthetic_data.py: voyage/vessel synthetic data with piracy and storm risk factors
Experimentation & Evaluation
4/10
How results are measured and tested
Basic but correct experimentation and evaluation: train/test split with stratification, ROC-AUC / AP / Brier reporting, and persisted metrics; no experiment tracking system observed.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/train_model.py: train_test_split, roc_auc_score, average_precision_score, brier_score_loss; metrics saved to data/model_metrics.json
AI-Risk-Scoring-for-Commercial-Insurance/marine_cargo/train_model.py: equivalent evaluation and metrics persistence
MLOps & Deployment
4/10
How models are shipped to production
Practical deployment artifacts: FastAPI endpoints and Streamlit frontend for NL2SQL, model persistence via joblib, Chroma vector ingestion and RAG-style retrieval; includes SQL AST safety check using sqlglot.
Evidence
NL2SQL-Datawarehouse-/main.py: FastAPI endpoints (/generate, /execute, /metadata/upload), OllamaEmbeddings + Chroma retrieval, sqlglot-based validate_sql_safety
AI-Risk-Scoring-for-Commercial-Insurance/train_model.py files: joblib.dump of trained models and model_columns.json for serving
Computational Efficiency
2/10
How efficiently computing resources are used
Minimal computational-efficiency engineering: sensible model hyperparameters and handling of class imbalance but no GPU/quantization/batching or profiling work.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/*/train_model.py: XGBClassifier parameters and use of scale_pos_weight to handle class imbalance
Research Depth & Innovation
1/10
Depth of research and new ideas
No evidence of novel research, custom architectures, or reproduced SOTA implementations; project is applied engineering rather than research innovation.
Evidence
train_model.py and feature_engineering.py patterns are applied and conventional; no custom layers or paper-implementation files present
Expertise
RAG• Middle
LLM• Middle
MLOps & Model Lifecycle• Middle
Industries
Artificial Intelligence• Middle
Financial Services• Middle
Transportation & Logistics• Middle
Technologies
Python• Senior • 5y+
SQL• Senior • 26y+
SAP HANA
Snowflake
LangGraph
AutoGen
Rest API
Spark
Databricks
Oracle• 26y+
Red Hat
VMWare
dbt
Reinforcement Learning• 5y+
Computer Vision
NLP• 7y+
Azure
MS SQL• 17y+
AWS
Ray
Azure SQL Database
Vector
Anomaly Detection• 7y+
Time Series Forecasting• 3y+
Replicate
Recommendations
- Develop production RAG-enabled NL2SQL services with hardened guardrails and monitoring (build on main.py generate_sql_with_rag_async and validate_sql_safety).
- Implement end-to-end insurance risk scoring inference services and batching logic using persisted model artifacts and FastAPI endpoints used in training scripts.
- Add experiment tracking and CI for model lifecycle (e.g., MLFlow/W&B integration, automated metric collection for the existing train scripts).
- Expand model monitoring and observability (prediction drift, calibration monitoring, and periodic re-training workflows) for deployed models.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Data Scientist
Confidence: Medium Data Engineer
Data engineer at a mid level specializing in end-to-end synthetic-data pipelines and batch feature engineering for insurance and risk scoring. The strongest proven skill is building realistic synthetic generators and batch feature rollups together with model training and artifactization, as shown by the telematics and marine cargo synthetic-data and training scripts. There is limited evidence of production-scale distributed systems, systematic statistical experimentation, or automated MLOps and data-versioning in public code.
Statistical Rigor
3/10
Correct use of statistics
Basic model evaluation metrics are present (ROC-AUC, average precision, Brier score) and class-imbalance handling is considered, but there is little statistical testing, uncertainty quantification, or explicit assumption validation.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/train_model.py: ROC-AUC, average_precision_score, brier_score_loss usage and scale_pos_weight handling
AI-Risk-Scoring-for-Commercial-Insurance/marine_cargo/train_model.py: same evaluation metrics and pos_weight calculation
Data Wrangling & Cleaning
6/10
Preparing and cleaning data
Thoughtful synthetic data generation, careful aggregation and null-handling, and explicit feature rollups show solid data wrangling and realistic dataset provenance for prototyping.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/generate_synthetic_data.py: RNG seeding and connected fleets/drivers/events/claims generation
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/feature_engineering.py: rolling aggregates, safe divide handling, merging drivers/fleets and claim label construction
Exploratory Analysis & Visualization
2/10
Exploring and visualizing data
Minimal exploratory analysis or narrative interpretation; a few printed summaries and feature-importance dumps exist but no structured EDA, plots, or written findings.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/feature_engineering.py: printed claim rate
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/train_model.py: printed top feature importances
Predictive Modeling
4/10
Building models that predict
Predictive modeling follows sensible engineering practice (stratified holdout, imbalance handling, XGBoost tuning and artifact persistence) but lacks cross-validation, calibration procedures, and detailed error analysis.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/train_model.py: stratified train_test_split, XGBClassifier hyperparameters, model save to joblib
AI-Risk-Scoring-for-Commercial-Insurance/marine_cargo/train_model.py: similar XGBoost training and evaluation
Business Insight & Impact
3/10
Turning analysis into business value
Domain-aligned features and comments reference regulatory explainability and operational scoring, indicating business awareness, but there is limited explicit mapping to business metrics or cost-of-error analysis in code.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/feature_engineering.py: domain-driven feature choices for telematics underwriting
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/train_model.py: comments on explainability and SHAP usage
Reproducibility & Notebook Hygiene
4/10
Clean, repeatable analysis
Reproducibility basics are present (seeded RNG, model random_state, saved model and column metadata, requirements files), but there is no evidence of CI, data/versioning, or containerized/pinned environment artifacts.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/generate_synthetic_data.py: RNG = np.random.default_rng(42)
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/train_model.py: model random_state and joblib.dump; requirements.txt present
Expertise
Analytics• Middle
Industries
Energy & Utilities• Middle
Financial Services• Middle
Transportation & Logistics• Middle
Technologies
LangChain
Chroma
SQLAlchemy
Power BI
XGBoost
FastAPI
Scikit-learn
Beautiful Soup
Pandas
NumPy
Streamlit
Pydantic
Uvicorn
Analytics• mentioned only
Recommendations
- Develop prototype risk-scoring pipelines that include synthetic-data generation, batch feature computation, and model training for insurance or logistics products.
- Implement production-ready batch ETL jobs or feature-store integrations (Airflow, dbt, or Feast) to operationalize the existing feature engineering code.
- Build internal NL2SQL tooling or lightweight model-serving endpoints that combine schema-RAG and SQL safety checks for enterprise data teams.
- Add robust evaluation workflows: cross-validation, calibration techniques, error-analysis dashboards, and data-versioning (DVC) for reproducibility
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Backend Developer
Confidence: Medium API Engineer
A backend-focused Python engineer (mid-level) who builds pragmatic ML-driven APIs and local LLM-backed tooling; the developer's primary strength is integrating ML/RAG pipelines into clear API surfaces. The strongest proven skill is implementing an NL2SQL RAG pipeline with SQL safety guards, shown by NL2SQL-Datawarehouse-/main.py's generate_sql_with_rag_async, sqlglot-based validate_sql_safety, and execute_sql_endpoint. What is not evidenced is production-grade ops maturity such as authenticated APIs, structured observability, migrations/DB schema evolution, retries/backoff, or comprehensive automated tests.
API Design
4/10
How well APIs are designed
Practical API design with FastAPI models, response schemas and consistent HTTP error handling; lacks versioned lifecycle, idempotency, pagination or advanced API governance.
Evidence
NL2SQL-Datawarehouse-/main.py: FastAPI app with /api/v1/nl2sql/generate and /api/v1/nl2sql/execute endpoints and response_model declarations
NL2SQL-Datawarehouse-/main.py: generate_sql_endpoint and execute_sql_endpoint use HTTPException for error contracts
NL2SQL-Datawarehouse-/app.py: Streamlit frontend calling the API and handling error responses
Data Layer & Database
3/10
Working with databases
Direct database usage with SQLAlchemy engine and SQL safety checks via AST parsing; no migration history, explicit transaction handling, or hand-tuned SQL optimizations are present.
Evidence
NL2SQL-Datawarehouse-/main.py: create_engine(connection_url, pool_pre_ping=True) and connection.execute(text(request.sql))
NL2SQL-Datawarehouse-/main.py: validate_sql_safety(sql) using sqlglot.parse_one(read='tsql') to enforce SELECT-only queries
Scalability & Performance
2/10
Handling load and speed
Some async/vector-search usage and minor DB pool configuration are present, but there is no caching/invalidation strategy, rate limiting, resilient retries or measured performance tuning.
Evidence
NL2SQL-Datawarehouse-/main.py: vector_store.asimilarity_search used asynchronously in generate_sql_with_rag_async
NL2SQL-Datawarehouse-/main.py: SQLAlchemy engine created with pool_pre_ping=True
System Architecture
3/10
Overall system structure
Clear separation between ML pipelines, feature-engineering and API surface shows pragmatic module boundaries; overall architecture is demo-oriented rather than production-grade microservice decomposition with service contracts and config management.
Evidence
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/feature_engineering.py: batch feature rollup and join logic separated from model training
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/train_model.py and NL2SQL-Datawarehouse-/main.py: separate modules for model training and API/RAG logic
Security & Auth
3/10
Protecting data and access
Security-aware choices include SQL AST validation and LLM output cleaning, but there are clear gaps such as no authentication on APIs and default hardcoded DB credential defaults present in code.
Evidence
NL2SQL-Datawarehouse-/main.py: validate_sql_safety using sqlglot and cleaned_sql = raw_sql.replace(...) removes code fences
NL2SQL-Datawarehouse-/main.py: MSSQL_PASSWORD default value present in environment fallback (hardcoded default)
Reliability & Observability
2/10
Stability and monitoring
Basic error handling and graceful failure responses exist, and ML training persists metrics; deeper reliability and observability practices like structured logging, correlation ids, retries with backoff, graceful shutdown, and metrics/alerts are not evidenced.
Evidence
NL2SQL-Datawarehouse-/main.py: execute_sql_endpoint wraps DB execution in try/except and raises HTTPException on errors
AI-Risk-Scoring-for-Commercial-Insurance/truck_telematics/train_model.py: model and metrics are persisted to disk (joblib.dump and model_metrics.json)
Expertise
Backend AI & LLM• Middle
Python• Middle
Databases & Vector Storage• Middle
Industries
Data & Analytics• Middle
Financial Services• Middle
Recommendations
- Productionize the NL2SQL API: add authentication, request rate limiting, structured logging with correlation ids, and retries with exponential backoff for external calls
- Implement database migration history and explicit transaction handling for any future write paths; remove hardcoded credential defaults and integrate a secrets manager
- Add tests and CI for the ML and API pipelines including unit tests for SQL safety validation and integration tests for the vector-store retrieval + LLM generation path
- Harden reliability: introduce metrics (Prometheus), tracing (OpenTelemetry/Jaeger), and graceful shutdown handling for async components
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: