AI & Software Engineer

Building Intelligent Agent Infrastructure

AI & Software Engineer specialising in agentic architectures, retrieval-augmented generation, and multimodal model optimisation, with experience building production AI systems and publishing technical articles for the developer community.

Research GitHub Twitter LinkedIn Medium
Agentic AI Systems LLM Infrastructure Multimodal Reasoning Generative Modelling
Phrugsa Limbunlom

The only limit is your imagination.

Highlight

Founder Live · Now in Beta

Receptr - AI Agent Platform for Non-Technical Businesses

Receptr is an AI platform designed to help non-technical founders deploy intelligent agents that can perform operational tasks such as customer support, marketing assistance, and business data analysis. The system focuses on simplifying AI adoption for small businesses by providing an orchestration layer that allows agents to understand business context and execute tasks autonomously. This project explores how agent architectures can reduce operational overhead for small companies.

Visit Receptr
Receptr” AI Agent Platform preview
01

About Me

My work focuses on building practical AI systems that combine reasoning models, structured retrieval, and autonomous agents to solve real-world problems. I have developed multiple LLM applications including research assistants, agentic product search systems, and scalable inference pipelines. I also publish technical articles exploring modern AI engineering topics such as GRPO training, LoRA fine-tuning, and production RAG architectures.

I hold an MSc in Artificial Intelligence (Distinction) from the University of Essex, where I worked on predictive modelling, generative modelling, and large language models. I currently work as an Analyst Developer at the University of Bedfordshire, contributing to AI-driven cloud systems that support the digital transformation of enterprise platforms. In parallel, I am a founder building AI agent infrastructure, an open-source contributor, and a technical writer publishing in-depth articles for leading AI publications.

Impact Highlights

  • Developed multiple LLM-powered systems including agentic research assistants and product search platforms
  • Published technical AI articles covering GRPO training, RAG pipelines, and multimodal fine-tuning
  • Built scalable LLM inference pipelines using AWS SageMaker and modern ML infrastructure
  • Experience deploying AI systems within enterprise environments including work with IT consultanting services
  • MSc Artificial Intelligence (Distinction) from University of Essex
MSc AI - UofEssex 6+ Certifications 10+ Projects 7+ Articles
02

Research Interests

  • Agentic AI systems
  • Multimodal reasoning architectures
  • Generative world models
  • Out-of-distribution generalisation

My research direction focuses on multimodal intelligence and world models as a path toward more robust and data-efficient machine learning systems. This direction was shaped by my earlier work, where I observed that many state-of-the-art models rely heavily on large-scale labeled datasets and often fail to generalise when faced with out-of-distribution scenarios.I am particularly interested in building predictive world models that capture the underlying structure, dynamics, and causal relationships of environments across modalities such as vision, language, and action.

03

Previous Research

Key research projects and contributions.

SE-GAN: Sentiment-Enhanced GAN for Stock Price Forecasting

(SeGAN)

SeGAN Architecture
SeGAN Results

Stock price prediction is a challenging task due to the inherent volatility of the market and the complexity of price movement. The proposed model, Sentiment-Enhanced GAN (SE-GAN), integrates sentiment analysis with generative adversarial networks (GANs) to generate more robust and accurate stock price predictions. The SE-GAN model demonstrates the lowest Root Mean Square Error (RMSE) compared to baselines including LSTM, GRU, and TimeGPT.

SeGAN comparison

Keywords: Stock Price Prediction, Sentiment Analysis, GANs, FINBERT

MANDY: Mandibular Fracture Classification & Localization on X-ray Images

(GradCAM)

MANDY

This research investigates relevant models for X-ray image classification using convolutional neural networks (CNNs) and transfer learning to identify mandibular fractures, and employs Gradient-weighted Class Activation Mapping (GradCAM) to localize fracture locations. The trained model was deployed as a diagnostic system to assist non-specialist doctors.

Keywords: CNN, GradCAM, Mandibular Fracture, Classification, Localization, X-ray

04

Paper

Sequential input and output prediction
Preprint

Deep Adversarial Learning for Financial Time-Series Forecasting (SE-GAN)

P. Limbunlom, TechRxiv, Sep. 9, 2025

Read Paper
05

Technical Publications

I regularly publish articles explaining AI-related topics including reasoning optimisation, multimodal model training, and retrieval-augmented generation. These articles focus on practical implementation techniques for modern AI systems and are read by developers interested in applied machine learning.

GRPO
Article

Training your reasoning model with GRPO

A practical guide for VLMs Post Training with TRL

Read Article
BitsAndBytes
Article

Choosing the right model size to your memory

Understanding bits & bytes and Model Quantization

Read Article
LoRA
Article

LoRA in Vision Language Models

Efficient Fine-tuning with LLaVA

Read Article
Show More
06

AI Systems & Platforms

I develop AI systems that combine retrieval-augmented generation, multi-agent reasoning, and scalable LLM inference to solve real-world problems. A project example is PickSmart (open source), an agentic AI product search platform that uses retrieval-augmented generation and LLM agents to analyse product reviews and generate personalised recommendations.

VLM GRPO
Project

VLM GRPO

Vision-Language Model Reasoning Optimisation with GRPO

View System
VLM LoRA
Project

VLM LoRA

Efficient Fine-Tuning Framework for Vision-Language Models

View System
SEAD-Agent
Project

SEAD-Agent

AI Research Assistant System with RAG and Contextual Reasoning

View System
Show More
07

Education & Credentials

Academic foundation and professional certifications.

Education

  • MSc, Artificial Intelligence and its Applications
    CSEE, University of Essex (Distinction)
    2023 - 2024
  • BSc, Major Computer Science
    Faculty of ICT, Mahidol University — 3.62/4.00
    2016 - 2020
  • Certificate, Artificial Intelligence (87%)
    ISS 2019, Sungkyunkwan University
    2019

Certifications

  • AWS Certified: Machine Learning Engineer Associate (MLA-C01)
  • Microsoft Certified: Azure AI Engineer Associate (AI-102)
  • IBM Generative AI Engineering with LLM Specialization
  • NVIDIA AI Infrastructure and Operations Fundamentals
  • Deep Learning Specialization
  • AWS Cloud Practitioner
AWS ML AWS Cloud Azure AI
08

Awards

09

More About Me

I am particularly interested in building AI systems that bridge the gap between cutting-edge research and real-world applications. My goal is to design infrastructure and platforms that make advanced AI capabilities accessible to organisations and developers.

Apart from technology, I enjoy science, evolution, and history. Visiting museums is another hobby of mine.