███╗ ██╗ █████╗ ████████╗████████╗██╗ ██╗ █████╗ ██╗ ██╗██╗███╗ ██╗
████╗ ██║██╔══██╗╚══██╔══╝╚══██╔══╝██║ ██║██╔══██╗██║ ██║██║████╗ ██║
██╔██╗ ██║███████║ ██║ ██║ ███████║███████║██║ █╗ ██║██║██╔██╗ ██║
██║╚██╗██║██╔══██║ ██║ ██║ ██╔══██║██╔══██║██║███╗██║██║██║╚██╗██║
██║ ╚████║██║ ██║ ██║ ██║ ██║ ██║██║ ██║╚███╔███╔╝██║██║ ╚████║
╚═╝ ╚═══╝╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝╚═╝ ╚═╝ ╚══╝╚══╝ ╚═╝╚═╝ ╚═══╝
AI & ML Engineer · Data Scientist · IoT Systems Engineer
Industrial Physics & IoT Engineering @ KMITL · Super AI Engineer S5 🥉 Bronze Medalist
📍 Pathum Thani, TH · 📧 natwin1313@gmail.com · 📱 098-832-5128 · 🏫 KMITL
class NatthawinTamprasert:
def __init__(self):
self.name = "Natthawin Tamprasert (ณัฐวินท์ แต้มประเสริฐ)"
self.role = "Data Scientist & AI Developer"
self.degree = "BSc Physics + BEng IoT Systems Engineering (Dual Degree) @ KMITL"
self.medal = "🥉 Bronze Medal — Super AI Engineer Season 5"
self.record = "Won 4 of 7 National Hackathons (AIAT · CPAXTRA · BOL · CHULA)
self.interests = [
"Time-Series Forecasting", "NLP", "Computer Vision",
"RAG / LLM Systems", "IoT + AI Integration"
]| Award | Event | Organizer | |
|---|---|---|---|
| 🥉 | Bronze Medal (Individual) | Super AI Engineer Season 5 | AIAT |
| 🥇 | 1st Place | Demand Forecasting Hackathon | AIAT × CPAXTRA |
| 🥇 | 1st Place | Crime Charges Analysis (NLP) | AIAT × BOL |
| 🥇 | 1st Place | Liver Fibrosis Severity Prediction | AIAT × CHULA |
| 🥈 | 1st Runner-up | Sales Forecasting & Product Optimization | AIAT × CPAXTRA |
| 🏅 | Honorable Mention | Crime Charges Analysis | AIAT |
| 🔟 | Top 10 Representative | SMART FACTORY Hackathon 2025 | Sony |
| 📜 | Certificate of Completion | AI Builders Program | AI Builders 2023 |
| 📜 | Certificate of Accomplishment | Brain Code 101 AI Builders Advanced | AI Builders 2023 |
Dhipaya Insurance (ทิพยประกันภัย) · 2025
Developed an explainable 4-stage Vision AI pipeline addressing 17.5% valuation discrepancy across 119,212 insurance claims
Stage 1 Car / Non-Car Filter CLIP Prompt Ensembling (zero-shot)
Stage 2 Damage Localization DINOv2 + Top-K Attention Pooling Hamming Acc: 0.84
Stage 3 Damage Classification ResNet-101 + DCN + FPN 20+ damage classes
Stage 4 Damage % Estimation Geometric formula: A_crop = A_full / Z²
└── Report Generation LLaMA 3.1-8B (4-bit) plain-language output for stakeholders
Impact — ↑ Transparency · ↑ Negotiation leverage · ↓ Claim costs
🥇 Demand Forecasting — 1st Place · AIAT × CPAXTRA · 19–23 May 2025
- Led comprehensive EDA revealing category-specific demand patterns; shifted strategy from global → category-wise modeling
- Fine-tuned Chronos pre-trained time-series models with custom post-processing
- Applied PCA + KMeans clustering to identify seasonal patterns and variability
🥇 Crime Charges Analysis (NLP) — 1st Place · AIAT × BOL · 12–16 May 2025
- End-to-end NLP pipeline with XLM-RoBERTa-Large for Thai legal document classification
- NER for bias reduction through name masking
- RAG system using FAISS + cosine similarity; BERT + LLM hardness-routing fallback for complex cases
🥇 Liver Fibrosis Severity Prediction — 1st Place · AIAT × CHULA · 2–6 Jun 2025
- Multi-modal AI combining ultrasound images + clinical tabular data
- AutoGluon + ResNetV2_50x for latent feature extraction
- Domain-aware post-processing → MAE improved 2.9 → 2.4
🥈 Sales Forecasting & Product Optimization — 1st Runner-up · AIAT × CPAXTRA · 5–9 May 2025
- Multi-tiered modeling: statistical methods for sparse products, ML for data-rich segments
- Novel signal-based demand movement classification
- Memory optimization via batch processing + stratified sampling on large retail datasets
📜 CAI Club Hackathon 2025: 7-11 × AI · CPALL × Creative AI Club · 2025
- Built LIFT (Language model Intelligence For Thai's 7-11) — a social listening system that monitors Thai news to predict product demand spikes at 7-Eleven stores
- Pipeline: news ingestion → LLM-based summarization & product category mapping → Typhoon API sentiment classification (positive/negative) → high-demand product flagging
- Estimated opportunity cost savings of ฿42.5M if preventing a 7-day stockout of a viral product (e.g., Dubai chocolate) driven by food influencer trends
- Partner: Wisesight (Thai social listening platform)
🏭 Smart Factory Hackathon — Supply Chain AI System · KMITL · 2025
- Developed end-to-end AI system for inventory status analysis, risk assessment, and automated procurement recommendations across a supply chain pipeline
- Applied GMM (Gaussian Mixture Model) unsupervised clustering to discover hidden patterns and anomalies in inventory without requiring labeled data
- Deployed Qwen 2.5 GGUF as a fully local LLM to generate human-readable summaries — preventing sensitive corporate data from leaving the organization
🌱 KMITL Innovation Expo 2025 — Carbon Credit System · KMITL · 2025
- Developed AI-powered carbon credit prediction system using Transformer/LSTM models integrated with IoT sensors (CH₄, CO₂) and NB-IoT for real-time environmental monitoring
- Created monthly forecasting model combining sensor data with machine learning for carbon credit trading applications
|
🏭 Supply Chain AI System
Inventory analysis + risk assessment + automated procurement. Applied GMM unsupervised clustering for anomaly detection. Deployed Qwen 2.5 GGUF fully on-premise — zero data leakage. |
🤖 NongAuanIoT — Multi-Agent RAG Chatbot
Multi-Agent RAG + Typhoon-S 8B Thai LLM · Fully CPU-based. Intent Classifier (cosine sim, threshold 0.6) routes to 3 agents: Academic · TCAS (DuckDuckGo) · General. |
|
🎙️ SpExNote — Intelligent Lecture Recording
IoT + AI pipeline: Silero-VAD → SpEx+ (fine-tuned 15.28h Thai audio) → ClearVoice → Thonburian-Whisper → Qwen-RAG. Addresses needs of 83.63% of dual-degree students. |
🔒 Smart Locker System
2× Arduino R4 WiFi + HC-SR04P + Micro Servo. M2M via MQTT + Node-RED Web API. Auto-password reset after 2 min of inactivity. |
|
🌿 Carbon Credit Prediction
Transformer/LSTM + IoT sensors (CH4, CO2) + NB-IoT for real-time environmental monitoring and carbon trading applications. |
🌊 Chao Phraya Water Quality
Benchmarked XGBoost · CatBoost · LSTM · GRU · Transformer with comprehensive metrics alongside ECE Master's students. |
Data Science & ML
LLM & Generative AI
Computer Vision
IoT & Hardware
DevOps & Tools
Languages
King Mongkut's Institute of Technology Ladkrabang (KMITL) BSc Industrial Physics + BEng IoT Systems Engineering · Dual Degree · 2025–Present
Chitralada School Science–Mathematics Program · 2020–2023 · Science GPA: 3.50
"The best model is the one that ships and makes impact."