Banyaphon Kongtham

bk.banyaphon@gmail.com, +66 631594914, Pathum Wan, Bangkok

Summary

Data Scientist with hands-on experience in credit risk modeling, decision systems, and machine learning pipelines. Skilled in transforming complex analytical models into business-driven insights. Experienced in Probability of Default (PD) modeling, OCR systems, and end-to-end ML deployment.

Education

Chulalongkorn UniversityExpected May 2026
B.Sc. in Computer Science

Work Experience

AI Engineer Intern | ClickNext Co., Ltd.Jan 2026 - Present
  • Fine-tuned an OCR model (Surya) for Thai handwriting recognition, reducing Character Error Rate (CER) from 60.95% to 19.98% (−67%), significantly improving accuracy for real-world document processing.
  • Developed a computer vision and automatic speech recognition pipeline to extract structured numerical data from live broadcast feeds.
  • Collaborated with cross-functional teams to deliver ML solutions aligned with business requirements.

Project

Video Anomaly Detection System using SlowFast and Adaptive MissionGNNAug 2025 - Mar 2026
  • Built a scalable multi-GPU feature extraction pipeline using SlowFast R50 (temporal dynamics) and ImageBind (visual semantics), generating high-dimensional fused representations (3,328 features) from surveillance video data.
  • Enhanced the MissionGNN architecture by introducing a learnable projection layer to align fused features with a knowledge graph embedding space, enabling structured anomaly reasoning.
  • Developed a weakly supervised training framework using decaying thresholds and anomaly-aware loss functions, achieving mAUC 0.8376 on the UCF-Crime dataset.
Risk-Adjusted Return Optimization Framework (Credit Risk Modeling)Jan 2026 - Feb 2026
  • Built an end-to-end credit risk modeling pipeline, including data cleaning, feature engineering, and exploratory analysis on lending datasets.
  • Developed a Probability of Default (PD) model using XGBoost with monotonic constraints aligned with financial risk logic.
  • Applied isotonic regression calibration to ensure reliable and interpretable risk scores.
  • Designed and optimized a credit decision engine by tuning Probability of Default (PD) thresholds, achieving $1.11M economic profit, 24.66% RAROC, and 56.3% approval rate at an optimal 9% PD threshold, under Basel III constraints.
  • Translated model outputs into business visualizations (risk vs. return, PD thresholds) to support lending strategy decisions.
Multi-Label Emotion Detection in Tweets (BERTweetMA)Jan 2025 - Feb 2025
  • Built a multi-label classification model using BERTweet with attention mechanisms on 38K tweets.
  • Designed emotion-specific attention blocks with sigmoid-based classification.
  • Achieved: F1-micro 0.728 | F1-macro 0.689 | Jaccard 0.699.
Task Tracking Web ApplicationMar 2024 - Aug 2025
  • Designed and developed a role based task management system for project tracking and cross-team collaboration.
  • Led frontend development and UX/UI design using NextJS to improve usability and workflow clarity.
  • Architected system structure including task status, role permissions, and project hierarchy to ensure scalable and maintainable design.

Technical Skills