Sivakorn Khundilokrattaya

About Me

Engineering Intelligence from Data

Balancing software engineering discipline with statistical machine learning and financial regulatory compliance.

// CANDIDATE PROFILE & QUANTITATIVE FOCUS

Computer Engineering @ SWU specializing in transparent, production-grade statistical ML.

Bridging statistical credit risk scoring, Basel-aligned probability of default, and scalable batch ETL data pipelines.

Credit Risk (WoE/IV) Explainable AI (SHAP) Data Pipelines (ETL)

// END-TO-END METHODOLOGY & PIPELINE DISCIPLINE

SYSTEM READY

01. Zero-Leakage

Ingestion & Audit

Raw schema validation, missing data audit, and strict temporal out-of-time test splitting.

02. Monotonic WoE

WoE Calibration

Weight of Evidence transformation, Information Value filtering, and PSI stability monitoring.

03. Stratified 5-Fold

Calibrated ML

LightGBM/XGBoost classification, probability calibration, and asymmetric loss optimization.

04. Sub-50ms Serving

Regulated Serving

Instance-level SHAP attributions, ECOA reason code generation, and FastAPI Docker endpoints.

// ACADEMIC FOUNDATION, COURSEWORK & GOVERNANCE

B.Eng. Class of 2026

Core Technical Coursework (Srinakharinwirot University):

Machine Learning Database Systems Operating Systems Statistical Methods for Engineers Data Structures & Algorithms Computer Networks Software Engineering

Clean Architecture & OOP

Adheres strictly to PEP-8 standards, modular object-oriented pipeline design, and reproducible containerized microservices.

Model Governance & Ethics

Zero black-box deployments without local feature attribution; full compliance with ECOA, FCRA, and SR 11-7 standards.

Git Monorepo Sync github.com/svkhun ↗

Continuous integration, automated linters, versioned experiment tracking, and clean commit hygiene.

// TECHNICAL ARSENAL

Skills & Technologies

A curated ecosystem of programming languages, statistical modeling frameworks, and scalable cloud data pipelines engineered for high reliability.

Core / Languages

Python (Expert) R C C++ Java SQL (Postgres / MySQL) Bash Scripting

Data Science & ML

LightGBM XGBoost Scikit-Learn SHAP (TreeExplainer) LIME OptBinning (WoE) Pandas NumPy SMOTE-Tomek

Data Engineering & Cloud

PostgreSQL MySQL Apache Spark Apache Airflow Docker • Compose FastAPI SQLAlchemy

Productivity & Analytics

Git Linux Environment Jupyter Lab RStudio Microsoft Excel Notion

Spoken Languages

Thai (Native) English (B1+ Professional / Oxford Test) ↗ Japanese (Elementary)

Python Core Language

SQL Relational Querying

C++ Systems & Memory

Java Enterprise Backend

R Statistical Modeling

XGBoost Asymmetric Loss

LightGBM Gradient Boosting

Scikit-Learn Pipelines & CV

SHAP TreeExplainer XAI

OptBinning Monotonic WoE

PostgreSQL ACID Relational

Docker Containerization

FastAPI Low-Latency Serving

Python Core Language

SQL Relational Querying

C++ Systems & Memory

Java Enterprise Backend

R Statistical Modeling

XGBoost Asymmetric Loss

LightGBM Gradient Boosting

Scikit-Learn Pipelines & CV

SHAP TreeExplainer XAI

OptBinning Monotonic WoE

PostgreSQL ACID Relational

Docker Containerization

FastAPI Low-Latency Serving

PostgreSQL Data Warehousing

MySQL RDBMS Architecture

Apache Spark Distributed ETL

Apache Airflow DAG Orchestration

Docker Containerization

FastAPI Sub-30ms Microservices

GISTDA Sphere Geospatial APIs

Sentinel-1 SAR Remote Sensing

PostgreSQL Data Warehousing

MySQL RDBMS Architecture

Apache Spark Distributed ETL

Apache Airflow DAG Orchestration

Docker Containerization

FastAPI Sub-30ms Microservices

GISTDA Sphere Geospatial APIs

Sentinel-1 SAR Remote Sensing

// PRODUCTION ARCHITECTURES

Featured Engineering Projects

End-to-end production pipelines bridging regulatory credit risk scoring, clinical frailty diagnostics, and automated manufacturing telemetry.

[ FinTech / Risk Analytics ]

Basel II/III • ECOA / FCRA Mapped

Enterprise Credit Risk Scoring & Explainable AI (XAI)

Production-grade banking intelligence platform designed to replace uninterpretable black-box ML with monotonic Weight of Evidence (WoE) scorecard calibration, gradient boosted trees, and legally defensible adverse action notices.

Monotonic Risk Binning: Discretized continuous financial metrics into monotonic risk bins using OptBinning and Information Value (IV) selection, ensuring linear relationship with log-odds of default.
Local SHAP TreeExplainer: Computed instance-level Shapley values (f(x) = E[f(x)] + ∑φ_j) to automatically generate legally required Adverse Action reason codes mandated by ECOA (12 C.F.R. § 1002.9) and FCRA.
Sub-50ms Serving Microservice: Containerized FastAPI inference endpoint with Pydantic contract validation, streaming real-time Probability of Default (PD) scoring under rigorous latency SLAs.

0.892 ROC-AUC (0.784 Gini)

PSI < 0.10 Zero Covariate Drift

42.6% KS Separation Stat

< 50ms Inference Latency

[ Financial Default Forecasting ]

Aihack 2025 National Finalist

Long Overdue Debtor (LOD) Prediction System

Enterprise predictive machine learning pipeline forecasting 90+ day default risks on real consumer loans. Developed under intense hackathon pressure and blind leaderboard testing for AIRA & AIFUL and Chulalongkorn Business School.

Trained on ~40,000 Real Records: Conducted strict out-of-time evaluation across 32,524 train / 8,619 test blind samples, tackling a severe 1:12 default imbalance without target leakage.
Temporal Delinquency Velocity Ratios: Engineered early vs. scheduled installment repayment acceleration indices, revolving limit exhaustion slopes, and cross-fold smoothed categorical encodings.
5-Fold Stratified Cross-Validation: Tuned LightGBM and XGBoost models via Bayesian Optuna search, yielding 0.894 ROC-AUC on public and private blind test splits to qualify as a National Finalist.

0.894 ROC-AUC (5-Fold CV)

~40,000 Consumer Records

PSI < 0.10 Scorecard Stability

Finalist National Tier

[ Industrial IoT / Lambda Data Engineering & Predictive Quality ]

Hybrid Lambda Architecture • WebSockets • XAI

Industrial IoT & OEE Manufacturing Data Platform

Enterprise-grade Hybrid Lambda data engineering architecture for discrete manufacturing lines. Integrates sub-second PLC edge telemetry streaming, physical Data Quality Gates, automated idempotent hourly OEE Data Mart aggregations, and TreeSHAP root-cause explainability for factory operators.

Edge Ingestion & Physical Quality Gate: Ingests concurrent multi-machine PLC sensor streams (cycle times, vibration RMS/Kurtosis harmonics, bearing temp, press force, motor current) with physical boundary gates rejecting $t \le 0$ signals and sub-second WebSocket broadcasting.
Idempotent Batch OEE Data Mart: Computes hourly Overall Equipment Effectiveness ($\text{OEE} = A \times P \times Q$) via CTEs with ON CONFLICT DO UPDATE deterministic upserts, full execution runtime auditing, and error telemetry in PostgreSQL 15.
Virtual Metrology & TreeSHAP XAI: Combines unsupervised Isolation Forest for continuous machine health scoring (0–100%) with cost-sensitive LightGBM for defect prediction, translating high-dimensional sensor spikes into actionable shopfloor root-cause attributions.

< 1.0s WebSocket Stream

100% Idempotent Upsert

OEE = A×P×Q Hourly Data Mart

TreeSHAP Root Cause XAI

[ HealthTech / Preventive Clinical AI ]

True Innovation Launchpad 2026

GuardianAI – Clinical Frailty & Fall Risk Intervention

Preventive healthcare diagnostic platform combining calibrated XGBoost models with Explainable AI (SHAP) for localized biometric factor attribution, serving sub-30ms medical predictions via FastAPI.

Multi-Biometric Frailty Risk Screening: Analyzes gait cadence dynamics, grip force attenuation, and posture telemetry to detect early geriatric frailty onset before acute falls occur.
Local SHAP Biometric Attributions: Translates complex model features into physician-interpretable diagnostic reasons (e.g. +14% risk from step cadence asymmetry, -8% from balanced BMI).
Streaming Clinical REST Endpoints: Asynchronous FastAPI microservice serving risk scores and personalized rehabilitation recommendations to interactive clinician dashboards.

< 30ms FastAPI Latency

Calibrated XGBoost Platt Scaling

Local SHAP Clinical Attributions

Docker Container Isolation

// HACKATHONS & COMPETITIONS

Competitions & Industry Challenges

Demonstrated machine learning modeling on enterprise datasets, operational AI challenges, explainable healthcare architectures, geospatial risk pipelines, and data-driven business strategy.

CDG Hackathon 2026

CDG Group • High-Impact Operational AI Challenge

Team GrandGuardianAI Jul 2026

Tackled advanced operational AI challenges focusing on high-stakes workflow automation, fault-tolerant decision pipelines, and edge system reliability under extreme operational constraints.

Operational AI System Reliability Mission-Critical ML Edge AI

Geospatial Intelligence for Resilience Hackathon 2026

GISTDA • KMITL • KMUTT • Remote Sensing & Disaster Resilience

Team Witsawa Tuapralat • Spatial Lead Jan 2026

30m Grid High-Resolution Flash Flood Probability Mapping

Sentinel-1 SAR Remote Sensing Radar Backscatter & Soil Moisture

GLO-30 DEM Multi-Source Slope, Flow & Precipitation Ingestion

Developed the mathematical framework and spatial ML pipeline fusing Sentinel-1 SAR backscatter, Copernicus DEM topography, and rainfall telemetry for 30m flash flood risk modeling.

Sentinel-1 SAR Copernicus DEM Spatial ML Flash Flood Prediction

True Innovation Launchpad 2026 – GuardianAI

True Corporation • Healthcare AI & Preventive Clinical Analytics

Team Nakphatthana Tuapralat • AI Lead Jan 2026

< 30ms Low-Latency FastAPI Model Serving

Calibrated XGBoost Multi-Biometric Frailty Risk Classification

Local SHAP Clinical Interpretability & Factor Attribution

Architected end-to-end clinical risk prediction platform for early elderly frailty intervention, combining calibrated XGBoost with SHAP biometric explainability for medical professionals.

Aihack Thailand 2025

AIRA & AIFUL • Chulalongkorn Business School • ProbSpace

National Finalist Dec 2025

~40,000 Real-World Consumer Credit Samples

0.894 ROC-AUC Top Tier Generalization on Blind Private Splits

Finalist Pitch Strategic P&L and Risk Trade-off Defense

Led ML and financial risk modeling to predict Long Overdue Debtors (>90 days default) on ~40,000 credit records using LightGBM/XGBoost feature velocity and calibrated decision curves under blind private split constraints.

LINE MAN Campus VIP Growth Campaign

Campus Business Strategy & Acquisition Analytics • LINE MAN Wongnai

Team Low Cortisol • Unit Economics Lead Nov 2024

8.87 THB Ultra-low Cost per Acquisition (CAC)

462 Students Acquired within Campus Hub (14-Day Trial)

8,000 THB Budget Cap • 100% On-Budget Execution

Analyzed university dining bottlenecks and user ordering behavior to formulate a data-driven LINE MAN VIP acquisition strategy, achieving 462 student conversions within a strict 8,000 THB budget cap at an optimized CAC of 8.87 THB.

Regulatory Framework (ECOA / FCRA / SR 11-7)

In regulated banking, models cannot operate as black boxes. The Equal Credit Opportunity Act (ECOA, 12 C.F.R. § 1002.9) and Fair Credit Reporting Act (FCRA) mandate automated, mathematically defensible Adverse Action Notices containing top adverse factors for rejected credit applicants.

Local vs. Global Interpretability (XAI)

Distinguishes global tree importance from instance-level Shapley values (f(x) = E[f(x)] + ∑_j=1^M φ_j). Integrates SHAP TreeExplainer to output exact marginal risk contributions for transparent credit underwriting.

Raw Credit Data

WoE / IV Binning

Calibrated LightGBM

SHAP Waterfall

Sub-30ms FastAPI

Weight of Evidence (WoE) & IV

Discretizes continuous variables into monotonic risk bins (WoE = ln(Good% / Bad%)). Enforces linear relationship with log-odds of default, automatically neutralizing outliers and capturing non-linear relationships.

Population Stability Index (PSI)

Monitors covariate shift across quarterly baseline vs. actual borrower distributions (PSI = ∑ (A_i - E_i) × ln(A_i / E_i)). Enforces PSI < 0.10 threshold to prevent uncontained feature drift.

Asymmetric Cost Utility Matrix

Underwriting optimization where False Negatives (uncontained write-off loss on principal, C_FN ≈ $10,000) cost ≈ 8× that of False Positives (forgone interest margin on rejected prime borrower, C_FP ≈ $1,200).

Applicant Risk Profiles:

Monthly Income (THB) 65,000 THB

Debt-to-Income / Utilization 32%

Past 24-Month Delinquency Count 0x

Predicted Probability of Default (PD):

PD: 8.2% Low Risk / Approved

Marginal SHAP Waterfall Contributions (ECOA / FCRA Adverse Reasons):

Baseline Discrimination: 0.894 ROC-AUC

5-Fold Stratified Cross-Validation on ~40k real consumer credit loan records.

Separation Metric: KS Statistic 42.6%

Clear separation between solvent and defaulting distributions at optimal cutoff τ = 0.35.

Population Stability: PSI < 0.10

Zero significant drift between training and Out-of-Time (OOT) test partitions.

Regulatory Compliance: Basel II/III & ECOA / FCRA

Fully mapped to Basel PD rating grades and Adverse Action Notice requirements.

Aihack Thailand 2025

Organized by AIRA & AIFUL Public Company Limited with Chulalongkorn Business School and ProbSpace at Jaiyossompati Building 3, Chulalongkorn University.

~40,000 Loan Records

Predicted default (>90 days overdue after 12 months) across 32,524 train / 8,619 test samples. Ingested temporal repayment velocity ratios, crunched 2 sudden blind test splits, and pitched to senior executives.

~40k Loan Samples

Temporal Repayment Lag

Velocity Slope Ratios

5-Fold Stratified CV

National Finalist Pitch

Payment Velocity Index

Ratio of early vs. scheduled installment repayments tracking debtor discipline.

Limit Exhaustion Slope

Rate of credit balance consumption over the initial 90-day onboarding window.

Temporal Delinquency Lag

Interaction between historical past-due cycles and payday monthly dates.

Smoothed Target Encoding

Branch and occupational risk categorical encoding with cross-fold smoothing.

• Tuned LightGBM (AUC: 0.894) • Baseline LogReg (AUC: 0.724)

Classification Cutoff (τ) 0.35

Sensitivity (Recall) 84.2%

Specificity 88.6%

Defaulters Caught (TP) 3,368

Net Benefit Generated $6.98M

Out-of-Fold (OOF) CV

Mean ROC-AUC: 0.8941 ± 0.0016
Evaluated with 5-Fold Stratified cross-validation, guaranteeing zero target leakage and resilient score stability across public and private ProbSpace splits.

Business P&L Trade-off

Solved the executive question sheet: optimized decision threshold to maximize cumulative loan interest margins while minimizing bad debt provision costs.

Preventive Healthcare Mission

Developed under True Innovation Launchpad 2026 (Team Nakphatthana Tuapralat). Evaluates multi-dimensional biometric telemetry to flag early-stage elderly frailty syndrome before acute deterioration occurs.

Actionable Diagnostic Insights

Translates complex machine learning risk predictions into localized clinical explanations, providing attending physicians with transparent risk attribution factors.

Biometric Telemetry

Feature Calibration

Calibrated XGBoost

SHAP TreeExplainer

Doctor Dashboard

Gait Velocity & Grip Dynamics

Tracks physical deterioration rates through sensor-derived cadence and force data.

Asymmetric Loss Penalty

Heavily penalizes false negatives to prevent missed opportunities for clinical intervention.

Platt Probability Scaling

Calibrates XGBoost raw margin outputs into true clinical posterior risk probabilities.

Localized Clinical Attribution

Integrates SHAP TreeExplainer to output granular clinical risk attributions (e.g. +14% risk from sudden step asymmetry, -8% from balanced BMI), empowering attending physicians with interpretable diagnostic reasoning.

Low-Latency Medical API

Deployed with asynchronous FastAPI endpoints streaming real-time frailty risk scores and individualized recommendations into an interactive React/Tailwind clinical doctor dashboard.

IIoT PLC Simulator

Physical DQ Gate

PostgreSQL 15 Time-Series

Speed & Batch Split

FastAPI WebSockets

React & Streamlit UI

1. Edge Simulator & Degradation Physics

Python simulator (simulator.py) emulating multi-machine PLC outputs across lines (LINE_01, LINE_02) and machine profiles (CNC_A, CNC_B, ROBOT_ARM). Calibrated to automotive parts manufacturing baselines (NHK Spring standard). Simulates cumulative tool wear (0.0 to 1.0), micro-cracks via vibration kurtosis (3.0 up to 14.0), bearing temperatures, press forces, and realistic fault transitions.

2. Relational Store & Dimensional Data Mart

PostgreSQL 15 relational store (mfg_postgres) configured with composite B-Tree indexes on (timestamp) and (line_id, machine_id). Houses raw telemetry (machine_telemetry), master downtime classification (downtime_reasons), aggregated hourly OEE Data Mart (hourly_production_summary), and operational audit logs (pipeline_execution_logs).

3. Speed Layer & Physical Quality Gate

Low-latency stream engine (stream_monitor.py) executing incremental watermark polling (MAX(event_id)). Enforces physical boundary checks intercepting non-positive cycle times ($t \le 0$) or corrupted sensor feeds, and tracks stateful consecutive defect streaks in-memory via collections.defaultdict(int) with compound keys to detect quality degradation spikes with zero database overhead.

OEE Mathematical Formulation ($\text{OEE} = A \times P \times Q$)

Overall Equipment Effectiveness (OEE) is the international discrete manufacturing benchmark evaluating productivity across three pillars: Availability ($A$), Performance ($P$), and Quality ($Q$). Computed dynamically in SQL CTEs over hourly time-slice windows.

Availability ($A$)

$\text{Run Time} / (\text{Run Time} + \text{Unplanned Downtime})$. Separates scheduled maintenance (Tool Change) from unscheduled breakdowns (Mechanical Jams, Sensor Faults).

Performance ($P$)

$\min(100.0, (\text{Ideal Cycle Time} \times \text{Produced Units}) / \text{Operating Time})$. Benchmarked against engineered machine limits: CNC_A (12s), CNC_B (15s), ROBOT_ARM (8s).

Quality ($Q$)

$\text{Good Units} / \text{Total Produced Units}$. Computes first-pass manufacturing yield per production cell and machine work center.

Idempotent Batch Upsert (PostgreSQL CTE)

Implements ON CONFLICT (hour_bucket, line_id, machine_id) DO UPDATE SET ... ensuring deterministic pipeline runs, historical backfills, and zero duplicate rows.

Pipeline Execution Audit Observability

Every batch job logs execution metadata (timestamps, duration, row counts, status, error stack traces) into pipeline_execution_logs for instant triage.

1. Unsupervised Machine Health Index

Isolation Forest model trained on normal operating baseline distributions without relying on historical defect labels. Converts decision function scores into an intuitive 0–100% Machine Health Index across 3 operational risk tiers: NORMAL ($\ge 75\%$), WARNING ($50-74\%$), and CRITICAL ($< 50\%$) to trigger proactive condition-based maintenance before catastrophic breakdown.

2. Virtual Metrology & Predictive Quality

Supervised LightGBM classifier predicting part defect probability before post-production quality assurance. Solves severe factory class imbalance via dynamic cost-sensitive loss weighting (scale_pos_weight = N_neg / N_pos) and calibrated decision threshold (0.35) prioritizing defect recall to prevent non-conforming parts from escaping to downstream assembly.

3. TreeSHAP Root Cause Attribution (XAI)

Instance-level Shapley value decomposition ($f(x) = \mathbb{E}[f(x)] + \sum_{j=1}^M \phi_j$) via TreeSHAP. Maps multi-sensor deviations (Vibration RMS, Kurtosis, Bearing Temp, Press Force, Motor Current, Crest Factor, Energy Proxy) into actionable Thai engineering descriptions (e.g. “อัตราเร่งแรงสั่นเทียบรอบก่อน”, “ดัชนี Crest Factor”) for shopfloor operators.

4. In-Memory Rolling Feature Extraction

Per-machine 10-cycle in-memory buffer (collections.deque(maxlen=10)) extracting 5-cycle rolling means, standard deviations, and lagged deltas ($\Delta t - \Delta t_{-1}$). Ensures zero train-serving skew by mirroring batch training transformations directly on streaming events with sub-millisecond latency.

FastAPI & Async WebSockets

High-concurrency backend (backend/) running on Uvicorn with connection pooling. Features a background broadcaster pushing real-time telemetry, machine health, and KPI updates to connected clients over /api/ws/telemetry.

React 18 + Vite Production UI

Modern responsive frontend (frontend/) with Tailwind CSS dark theme, glowing machine status beacons, animated WebSocket table feeds, Recharts OEE trends, and Web Audio API alarm sirens.

Containerized Microservices

Multi-service deployment orchestrated via Docker Compose: PostgreSQL 15, FastAPI backend, Nginx-proxied React UI, and legacy Streamlit BI dashboard (dashboard.py) with Plotly Express.

[SPEED] 2026-09-09 17:30:00 | Watermark @ event_id #14,892 | Ingested: 6 events across LINE_01, LINE_02

[DQG-PASS] Physical Quality Gate: 6/6 signals nominal (cycle_time > 0s, vibration RMS < 4.5mm/s)

[ML-HEALTH] Machine CNC_A | Health Index: 93.2% (NORMAL) | Defect Prob: 0.012 | Top Factor: Vibration RMS (+0.002)

[ETL-BATCH] Pipeline 'hourly_oee_batch_aggregation' executed in 14.8ms | Upserted 6 records into Data Mart

[OEE-SUMMARY] LINE_01 OEE: 89.4% (Avail: 94.8%, Perf: 97.2%, Qual: 97.0%) | Target: >= 85.0% [TARGET ACHIEVED]

[WEBSOCKET] Broadcasted telemetry & ML health snapshot to active client sessions

Peak Rush Dining Bottlenecks

Field research identified intense congestion in university dining halls during the 11:30–13:30 peak window, resulting in 25–35 min queues and fragmented peer orders with surge delivery fees.

Campus VIP Subscription Framing

Formulated the "Silent Mission" growth campaign positioning LINE MAN VIP benefits (zero delivery fees, group cart discounts, and synchronized batch dispatch windows) tailored specifically to campus dormitory clusters.

Cafeteria Bottleneck

Ambassador Outreach

14-Day VIP Activation

Group Cart Order

LTV Retention

Top-of-Funnel Ambassador Outreach

Engaged 1,800+ student foot-traffic impressions across campus faculty quads and study lounges.

462 Verified Student Conversions

Direct onboarding into LINE MAN VIP trial subscriptions via single-tap QR verification with 25.6% conversion.

Group Cart Basket Aggregation

Incentivized 3+ student pooling per delivery drop-off to reduce single-order delivery friction by 64%.

CAC Financial Architecture

Operated under a strict 8,000 THB budget cap. Total promotional collateral, ambassador stipends, and acquisition incentives resulted in an optimized net CAC of 8.87 THB per student, outperforming industry delivery benchmarks.

Unit Contribution Margin

Targeted break-even on the 2nd repeat order per subscriber. With average order value (AOV) at 160 THB and 15% platform take-rate, each retained subscriber returned 24 THB gross margin, covering acquisition cost within 10 days.

Day 7 & Day 14 Retention Hooks

Automated LINE push notifications triggered specifically during exam periods and evening study sessions.

Repeat Purchase Velocity

Tracked campus cohort order re-engagement rates, identifying a 2.4x ordering frequency lift among VIP subscribers.

Competition Jury Defense

Defended growth hypothesis, unit economics math, and scaling playbook before LINE MAN Wongnai executive jury.

Section	Score	CEFR	Time	Communicative Competencies
Use of English	54	B1	18:28	Uses sufficient range of language to explain main points in ideas or problems with reasonable precision. Demonstrates good vocabulary related to familiar topics and everyday situations. Demonstrates reasonable grammatical control with elementary vocabulary.
Listening	48	B1	35:00	Understands straightforward factual information about everyday and work-related topics. Follows conversation and discussion, including general themes and main points. Understands main points in various types of audio recordings delivered clearly.

Subject Domain	Score	Benchmark
1. Foundations of Data Analysis	20.00 / 30	18.87
2. Data Cleaning and Preprocessing	23.00 / 30	22.52
3. Exploratory Data Analysis and Visualization	20.00 / 30	16.23
4. Data Analysis Techniques and Statistical Methods	16.00 / 30	17.83
5. Machine Learning for Data Analysts	18.00 / 30	15.28
Total Score	97 / 150	90

Level	Score Range	Proficiency Standard
Beginner	0–60	Needs Improvement: Understands foundational concepts but lacks applied execution.
Intermediate • [Current Level: 97]	61–105	Adequate Foundation: Capable of intermediate execution across core workflows.
Proficient	106–135	Solid Proficiency: Demonstrates strong practical execution and efficient domain reasoning.
Expert	136–150	Advanced Mastery: Capable of resolving complex domain problems and high-level strategy.

Subject Domain	Score	Benchmark
1. Data-Driven Marketing	21.00 / 30	20.93
2. Customer Segmentation and Targeting	16.00 / 30	17.30
3. Digital Marketing Analytics [Top Performer]	26.00 / 30	17.94
4. Data-Driven Content Marketing	19.00 / 30	21.63
5. Marketing Strategy and Optimization	16.00 / 30	16.82
Total Score	98 / 150	94

Level	Score Range	Proficiency Standard
Beginner	0–60	Needs Improvement: Understands foundational concepts but lacks applied execution.
Intermediate • [Current Level: 98]	61–105	Adequate Foundation: Capable of intermediate execution across core workflows.
Proficient	106–135	Solid Proficiency: Demonstrates strong practical execution and efficient domain reasoning.
Expert	136–150	Advanced Mastery: Capable of resolving complex domain problems and high-level strategy.

Engineering Intelligence from Data

Computer Engineering @ SWU specializing in transparent, production-grade statistical ML.

Ingestion & Audit

WoE Calibration

Calibrated ML

Regulated Serving

Skills & Technologies

Featured Engineering Projects

Enterprise Credit Risk Scoring & Explainable AI (XAI)

Long Overdue Debtor (LOD) Prediction System

Industrial IoT & OEE Manufacturing Data Platform

GuardianAI – Clinical Frailty & Fall Risk Intervention

Competitions & Industry Challenges

CDG Hackathon 2026

Geospatial Intelligence for Resilience Hackathon 2026

True Innovation Launchpad 2026 – GuardianAI

Aihack Thailand 2025

LINE MAN Campus VIP Growth Campaign

Industry Certifications & Assessments

Data Analyst Assessment

Digital & Performance Marketer

Oxford Placement Test

Let's Connect

Sivakorn
Khundilokrattaya