Computer Engineering · Chulalongkorn

Nawat
Pimolsathien

นวัต พิมลเสถียร · วิศวกรคอมพิวเตอร์ (หมูแดง)

Computer Engineer · also known as Moodaeng 🐗

moodaeng@chula:~$ building AI things · shipping web apps · petting cats

Scroll to orbit

Portrait of Nawat Pimolsathien (นวัต พิมลเสถียร) ~ 4th year
01 — About

Following the rabbit holes, one stack at a time.

I'm a fourth-year Computer Engineering & Digital Technology student at Chulalongkorn University. Programming started as coursework and side projects, then turned into web development, SQL and data work through my internships, and lately machine learning, data analytics, IoT, cloud, and shipping desktop software and add-ins in C# and .NET. I like building things that are genuinely useful to the people who'll end up using them — ideally with a cat nearby and coffee within reach.

สวัสดีครับ ผม นวัต พิมลเสถียร (หมูแดง) นิสิตชั้นปีที่ 4 ภาควิชาวิศวกรรมคอมพิวเตอร์และเทคโนโลยีดิจิทัล (CEDT) คณะวิศวกรรมศาสตร์ จุฬาลงกรณ์มหาวิทยาลัย ปัจจุบันเป็น Software Engineer ที่ บริษัท เมืองไทยประกันชีวิต จำกัด (มหาชน) งานที่ทำอยู่ครอบคลุมตั้งแต่การพัฒนาเว็บด้วย React และ Next.js, ระบบฝั่งเซิร์ฟเวอร์และซอฟต์แวร์เดสก์ท็อปด้วย C# / .NET, ไปจนถึงงาน Machine Learning, Data Analytics, IoT และ Cloud ผมชอบงานที่ท้าทาย สร้างสรรค์ และมีประโยชน์ — ถ้ามีกาแฟด้วยจะดีมากครับ

02 — Experience

Where I've been building

Aug 2025 — Present

Software Engineer InternNOW

Muang Thai Life Assurance
  • First 9 months: built a sales-assistance chatbot using a multi-agent + RAG architecture with Python, LangChain and LangGraph.
  • Routed questions with a TF-IDF + logistic-regression intent classifier (scikit-learn), and evaluated retrieval with hit-rate and MRR against reference answers from a larger model, combined with human scoring.
  • Since then: developing and modernizing web applications migrated from legacy systems, writing unit and integration tests to guarantee behavior is preserved through the rewrite.
View my senior project poster (CCES2026) CCES2026 poster for the MTL Agency Chatbot senior project
Open full-size PDF ↗
May 2025 — Sep 2025

Software Developer (Intern / Part-time)

Power Partner Co., Ltd.
  • Developed add-ins for CAD software using C# and .NET.
  • Built new features from real client requirements.
  • Migrated existing add-ins from .NET Framework to .NET (Core) LTS.
See an example: PEstimate, a CAD add-in PEstimate ribbon toolbar inside ZWCAD, with tools for quantity takeoff, floor boundary, scale, and reports
A construction material cost-estimation add-in for ZWCAD — takes off quantities from a drawing (walls, columns, doors, floor area, etc.) and reports estimated material costs. It's an internal company product, so I can't share the source, but here's its ribbon toolbar.
A weekend project from this: AutoCAD + WebView2 add-in Concept diagram: an Open WebView button in a new hehe ribbon tab in AutoCAD, docking a WebView2 browser panel on top of the drawing
After working with CAD add-ins here, I built a small side project for fun: a WebView2 browser panel embedded inside AutoCAD — a new "hehe" ribbon tab with an "Open WebView" button that docks a live browser window on top of the drawing, so you don't have to alt-tab out to look things up. (Diagram above, since I don't have AutoCAD here to grab a real screenshot.)
Aug 2024 — Mar 2026

Teaching Assistant

Chulalongkorn University · Digital Computer Logic, Data Structures & Algorithms, Frontend Development
  • Focused on Frontend Development (Next.js) and Data Structures & Algorithms.
  • Debugged students' code daily, quickly finding the root cause of bugs across frontend and algorithm assignments.
  • Answered technical questions and graded homework.
  • Returned as Frontend Development TA for a second term (Mar 2026).
May 2024 — Jul 2024

Software Engineer Intern

NITMX Co., Ltd.
  • Built full-stack websites with Next.js and Node.js, and wrote backend code in Java, at Thailand's national payment infrastructure operator.
  • Designed UI in Figma and designed the database schema.
  • Studied the team's Go microservices and Kafka-based event streaming for payment processing.
03 — Projects

Built on my own time, measured not assumed

Machine learning · Python · QLoRA · GGUF

pholama — compressing a Thai LLM

Pruned and quantized Typhoon2 3B (Llama 3.2 base) into a domain Q&A model for Chulalongkorn's registrar office: vocabulary trimmed 128k → 45k, layers pruned 28 → 24, quality recovered with QLoRA (perplexity 14 → 7.6), then quantized to GGUF Q4_K_M and deployed behind retrieval. Every stage ran on a 4 GB laptop GPU.

  • −76%deploy size, 6.4 → 1.5 GB
  • 23.5 tok/son CPU vs 10.6 on GPU
  • 35 → 47.5%in-domain quality
GitHub ↗
Backend · Go · Kafka · PostgreSQL · Redis

satang — event-driven double-entry ledger

Money-transfer backend: idempotent transfers (Redis plus a database unique constraint), ordered row locking to prevent deadlocks, a transactional outbox to Kafka, and an async fraud-detection consumer with retries and a dead-letter queue. REST (Gin) and gRPC share one service layer; tested with testify and testcontainers, traced with OpenTelemetry and Jaeger, metrics in Prometheus and Grafana, shipped with Docker Compose, Kubernetes manifests and GitHub Actions CI.

GitHub ↗Live demo ↗
Realtime · Go · Redpanda (Kafka) · WebSocket · Svelte

OddsPulse — realtime prediction-market prices

Polymarket prices stream through a Redpanda pipeline into a Go WebSocket gateway that fans them out live to a browser Market Wall, Surge Feed and Top Movers panel, with a replay mode so the screen always moves.

  • 20,000concurrent WebSocket clients per gateway
  • 2×over the 10k target
GitHub ↗Live ↗
  • MooGPTGPT-style transformer written from scratch in C++ — tensors, backprop, AdamW, BPE tokenizer — every gradient checked against a PyTorch oracle.
  • typhoon_finetuneQLoRA fine-tune of Typhoon2-3B on a Thai speech-style dataset built from YouTube transcripts, trained inside 4 GB of VRAM.
  • air-hockey RLHybrid physics + reinforcement-learning bot (SAC, self-play) that decides at 10 Hz. Playable in the browser.
  • AurumOrder management for a gold retailer — Next.js + NestJS, PostgreSQL, Redis, MinIO, behind Caddy and a Cloudflare Tunnel. Private client work; walkthrough on request.
  • OpenGL CrackProcedural crack patterns on surfaces using Voronoi and parallax techniques, built with a collaborator.
  • Tecto StudioInteractive 3D tectonic-plates globe in Next.js: replay plate reconstructions from up to 1.8 billion years ago, or drag and slice plates in a sandbox. Runs fully in the browser.
All repositories on GitHub ↗
04 — Toolkit

Tech I reach for

// frontend

  • Next.js
  • React
  • Figma

// backend

  • .NET / C#
  • Java Spring Boot
  • Go
  • Node.js
  • NestJS
  • C++

// ai & ml

  • TensorFlow
  • scikit-learn
  • Python
  • LLMs
  • PyTorch
  • Hugging Face
  • PEFT / QLoRA
  • LangChain / LangGraph
  • PyThaiNLP
  • Gymnasium (RL)

// data & devops

  • PostgreSQL
  • SQL
  • Docker
  • Redis
  • MinIO
  • pandas
  • NumPy
  • Caddy
  • Cloudflare

// distributed systems

  • Kafka
  • gRPC
  • Kubernetes
  • Prometheus / Grafana
  • OpenTelemetry

// also into

  • Blender
  • Data Analytics
  • IoT
  • Cloud
05 — Off the clock

Education & the rest of me

B.Eng. Computer Engineering & Digital Technology
Chulalongkorn University
วิศวกรรมคอมพิวเตอร์และเทคโนโลยีดิจิทัล · จุฬาลงกรณ์มหาวิทยาลัย
2023 — 2026 · Year 4 · GPAX 3.71 / 4.00 · transcript ↗

Coursework: Software Engineering · Database Systems · Data Structures & Algorithms · Computer Networks · Operating Systems & Systems Programming · Neural Networks (A) · AI & Machine Learning · Introduction to Data Science Engineering (A) · Statistics for Computer Engineering · IoT Systems · Capstone Project I–II (A)

🇹🇭 Thai (ไทย) — native 🇬🇧 English — TOEIC 900/990 (B2) ↗