• Blog
• Hugging Face
• LinkedIn
• PDF

• Wangchanberta-Thainer-Corpus-v2-2 No.1 Download in AI-Research Apply Model in Thailand Fine-tuned airesearch/wangchanberta-base-att-spm-uncased on the pythainlp/thainer-corpus-v2.2 dataset.

  

  # Load model directly
  from transformers import AutoTokenizer, AutoModelForTokenClassification
  
  tokenizer = AutoTokenizer.from_pretrained("Porameht/wangchanberta-thainer-corpus-v2-2")
  model = AutoModelForTokenClassification.from_pretrained("Porameht/wangchanberta-thainer-corpus-v2-2")

• Object Detection API A FastAPI-based microservice for object detection tasks with asynchronous processing.

• ML Worker for object detection at scale RF-DETR Model ML Worker for object detection at scale. Pulls tasks from Pub/Sub, reads images from GCS, runs RFDETR, stores results to GCS, and notifies an internal API.

  

• Machine learning system design for realtime object detection

  

• Thongna High-performance text processing library for the Thai language, built with Rust and exposed as a Python package.

  

• Restaurant Recommendation API - Rust High-performance multi-threaded application restaurant recommendation service with ML inference, achieving P95 latency < 20ms at 800 concurrent users.

  

• BiLSTM-CRF using Word/Character Embedding Model for Named Entity Recognition (NER) Task—Address Detection and Classification in Thailand.

• Ecommerce Hybrid Search Utilizes Pinecone vector DB, CLIP dense vectors, and BM25 sparse vectors with hybrid scaling for product search.

  

• OpenThaiGPT-7B Customer Support Thai customer support fine-tuned version of the OpenThaiGPT-7B model.

• BERT Intent Classification in Thai Smart AI chat that classifies customer intent using BERT.

• Whisper-Tiny Smarthome in Thai Whisper-tiny fine-tuned for Thai smarthome speech, achieving 24.375 WER on eval set.

• Product Description in Thai Dataset 114k product descriptions in Thai.

• Booking Intent in Thai Dataset 10k synthetic Thai booking intent samples.

• rust-remove-background-cpu High Performance a fast background removal service written in Rust, using the U2Net model and ONNX Runtime for CPU Base inference.

  

• bg-remove-webgpu Image background removal powered by AI. Uses WebGPU acceleration for maximum speed, processes everything locally in your browser.

  

• image-processing-cpu A FastAPI service that removes backgrounds from images using deep learning models, optimized for CPU and low-memory environments.

• data-analyst-agent Multi-agent AI system with ease, flexible framework provided by crewAI. This agents to collaborate effectively on complex tasks

  

• Mechanical Engineer (⚙️ 5+ years)
  • Project owner and design product pipe tube seal, this is the special part of machine honda CITY 2020 Turbo (Major Change 3 Pistons)
  • Root Cause Analysis & Problem Solving expert for mechanical and electrical issues, with CNC programming expertise, sometimes will research chemical & material for finding root cause.
  • Build and develop search system ~500 records problem unique to match problems with their root causes using spreadsheets. this project for reduced the time spent investigating repetitive causes from ~30 minute sometimes up to 3 days to approximately 1 minute if the problem appears in the table!
  • Maintenance Department leader and advisor managing 30+ technicians with 400+ machines, main focus breakdown maintenance and i'm success in reduced breakdown time from ~7% to ~1% per month in industrial operations.
• Software Engineer (Full Stack) (💻 3+ years)
  • Co-Founder and developer in NFT Minting digital product 3000 pcs. on ETH Chain I'm focus in smart contract research, metadata storage IPFS for images and web full-stack development this project earn 25000$.
  • Improvement reduced time and space complexity of status change of campagin event from O(n) to O(1) by using redis ttl.
  • Maintainer Core Machine Learning model for Intent Thai&Eng Classification using the concept of contextual string embeddings which is used for Sequence Labelling.
  • Build and develop authentication system for telegram bot by utilization TON Chain Web3 Wallet Technology.
• Machine Learning Engineer (💻 2+ years -> past)
  • R&D and deploy NER Task using Deep Learning BiLSTM + CRF Algorithms for address extraction feature in Thai langauge.
  • Develop and maintainer public model on huggingface in NER Task using BERT with ThaiNER corpus 2.2
  • Optimize search system for reduced response time from ~800ms to ~200ms in hybrid search (dense + sprase) in qdrant vectordb feature 150k+ products large payload and images.
  • Research and optimized feature remove background U2NET Model from GPU-based to CPU-based using quantized models technique for reduced from ~800 mb to ~40 mb with Rust and ONNX Runtime.
• Passionate about building scalable systems, intelligent applications, and efficient microcontrollers
• Always learning, always tinkering—bridging the gap between hardware and high-level computation

King Mongkut's Institute of Technology Ladkrabang Bachelor of Engineering, ⚙️ Mechanical Engineering (2017)

Programming Languages: Python (Everyday), JavaScript (Everyday), Rust (Every Hobby project), C++ (Learning)

Web Development: React, Next.js, Astro, Node.js, Nest.js, Actix Web, FastAPI, Tokio, Axum, Tailwind CSS

AI & ML: TensorFlow, PyTorch, Hugging Face, Jupyter Notebook, DVC, Weights & Biases, MLFlow

Tools & Technologies: Docker, Kubernetes, Git, MongoDB, Redis, PostgreSQL, GitHub Actions, Jenkins, ArgoCD, CircleCI, AWS, GCP

"Always exploring, always building—at the crossroads of language, computation, and embedded systems."