Porameht Khumsombat
Software Engineer (AI/ML)
huggingface.co/poramehtAbout Me
Experienced in scalable system architecture and end-to-end AI/ML pipeline development. Extensive work with microservices design, NLP systems, and production deployment at scale. Active open-source maintainer with Thai NLP tools and models used by the community. Prolific technical writer with 100+ articles. Strong focus on zero-cost abstraction and performance optimization. Proficient in Python, Rust, TypeScript/JavaScript, React/Node.js, and cloud infrastructure.
Experience
Software Engineer (AI/ML)
- Design and architect scalable backend systems with microservices architecture and SOLID principles
- Develop multi-agent AI systems with distributed architecture patterns
- Build production-grade ML backend infrastructure with FastAPI and async programming patterns
- Implement software engineering best practices for maintainable and testable AI/ML systems
Machine Learning Engineer II
- Designed hybrid vector search system (dense + sparse) with Qdrant, batch and real-time processing across 1m+ skus and large payload
- Design and develop inference embedding model for search and ranking model with BentoML and MLFlow registry
- Optimize CDC online feature store real-time processing for ranking model ingest data to qdrant and algolia from ~7 sec. to ~1 sec. per record
- Built image editing tool utilizing gpt-image-1 model and GPT-4.1 nano, product No.37 of creative product award 2025 by Adaddict
Middle Software Engineer (AI/ML)
- Developed and deployed BiLSTM-CRF model for Thai NER address detection deployed to production with response time less than 200ms
- Research and optimized feature to 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.
- Designed and Optimize hybrid vector search system (dense + sparse) with Qdrant, enabling real-time semantic search across 150k+ products from ~800ms to ~200ms
- Architected scalable ML pipeline with W&B tracking and automated deployment by DVC to GKE, reducing model deployment time by 60%
Software Engineer (Full-stack)
- Optimized core feature status change of campaign to reduce time and space complexity from O(n) to O(1) by using redis
- Designed scalable microservices architecture using SOLID principles and DDD patterns with Kubernetes on GCP
- Developed SSR web application with SEO optimization and Open Graph protocol
- Set up automated CI/CD pipeline with Jenkins and GitOps workflow
Projects Open Source
-
A fine-tuned version of airesearch/wangchanberta-base-att-spm-uncased on an pythainlp/thainer-corpus-v2.2 dataset
-
High-performance Text processing library for the Thai language, built with Rust and exposed as a Python package.
-
Model for Named Entity Recognition (NER) Task Address Detection and Classification in Thailand.
-
Utilize Vector database Pinecone Dense Vectors CLIP + Sparse Vectors BM25 Hybrid vector scaling using a convex combination
-
A fine-tuned version of OpenThaiGPT-7B model on the Porameht/customer-support-th-26.9k dataset.
-
Smart AI chat that can classify the intent of customers using BERT (LLM ModeL) intent classification model.
-
Model is a fine-tuned version of openai/whisper-tiny on the processed-smarthome-th dataset. It achieves the following results on the evaluation set: Wer: 24.375
-
Background removal using WebGPU. and Transformer.js Inference Model on browser based.
-
A dataset of 114k product descriptions in Thai language.
-
A dataset (synthetic) of 10k booking intent in Thai language.
Education
-
King Mongkut's Institute of Technology Ladkrabang
Bachelor of Engineering
2017
Mechanical Engineering
Skills
-
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 Data Version Control, Weights & Biases, MLFlow, Prefect, Qdrant
-
Tools & Technologies
Docker, Kubernetes, Git, MongoDB, Redis, PostgreSQL, GITHUB Actions, Jenkins, ArgoCD, CircleCI, AWS Cloud, GCP Cloud
Certifications
-
Building Recommender Systems with Machine Learning and AI
LinkedIn Learning
Machine Learning, AI, Recommender Systems
-
LLM101x: Large Language Models: Application through Production
Databricks
Large Language Models, AI, Production Grade, LLMops
-
Classify Images with TensorFlow on Google Cloud
Google Cloud
TensorFlow, Cloud Computing
-
Data Science
Srinakharinwirot University
Data Science, Machine Learning, Language Modeling