Software Engineering · RMUTL Chiang Mai, Thailand

I build software that holds up in production: real-time apps, distributed backends and practical ML.

I'm Tawan Tiphkun, a software engineering student. I like taking a real problem, such as a restaurant's paper order slips, a trader who can't stop chasing losses or a production line counting beans by eye, and shipping something that solves it end to end.

See my work GitHub ↗ Email me

0.91: mAP@50 on my custom YOLO detector
5: architectures of one system, from layered to microservices
3: real-time apps sharing one BizFlow backend

01 Selected work

Full-stack · Team project

BizFlow: QR table ordering

Diners scan a QR code at their table, order from their phone and pay with a PromptPay QR. The kitchen and cashier see every order change instantly.

Prices are always recomputed on the server, so a tampered request can't change the bill
JWT roles (admin, kitchen, cashier) guard every endpoint
Orders, payments and table state change together in one database transaction
Socket.IO rooms push live order status to the kitchen and to each diner

TypeScript · React · Express · Prisma · MySQL · Socket.IO · Docker

View the code ↗

Mobile · Flutter

Tipkhun Capital

Most small traders don't lose money on one bad trade. They lose it by not stopping. This app turns a risk plan into hard numbers: a daily loss budget, a trade journal, an automatic STOP state and rules for splitting profit.

Flutter · Dart · iOS · Android · Web

View the code ↗

Tipkhun Capital overview showing the remaining daily risk budget

Machine learning · Computer vision

Bean snack detection with YOLO

A custom detector that finds and classifies the 5 ingredients of a Thai mixed-bean snack in photos, video and a live camera feed. I built every stage myself: data collection, labelling, dataset tooling, training, evaluation and inference.

0.914 mAP@50 and 0.93 recall on held-out images
About 120 ms per image on a CPU, with no GPU

Python · Ultralytics YOLO · PyTorch · OpenCV · Label Studio

View the code ↗

Model detecting peanuts, broad beans and green peas

Backend · Security

Secure Task Board

Microservices behind an Nginx gateway with JWT and RBAC, rate limiting, a database per service, and Grafana + Loki logging.

Node.js · PostgreSQL · Nginx · Docker

Database design

RPG Game Database

Game rules enforced with constraints and triggers, plus an admin panel with role-based access, bans and an audit log.

MySQL · Express · EJS

Real-time backend

Agent Wallboard

Call-center backend that pushes every agent status change to dashboards over WebSockets. Joi validation, Jest tests and CI.

Express · MongoDB · Socket.IO · Jest

Desktop

Secure Electron IPC

A sandboxed renderer, a minimal contextBridge API and allow-list validation in the main process.

Electron · JavaScript

Full-stack

User Management Admin

React admin panel and layered API with admin-only access, soft delete and real-time agent presence.

React · Express · SQLite · JWT

Full-stack

Restaurant Review App

Debounced live search, combinable filters, and reviews that are validated on the client and the server.

React · Vite · Express

02 Startup pitch: LandOpt

Beyond class projects, my team of three software engineering students took a product idea through RMUTL's University Business Incubator (UBI) startup programme in Chiang Mai, from problem research and a UI prototype to a five-minute pitch in front of a judging panel (February 2026).

Our team pitching 'AI land tax assessment and vacant-land management' on stage, with the 5-minute timer running — Pitching on stage. Every team had five minutes, then questions from the judges.

PropTech · AI · Business model

TaxSave.AI: turning vacant-land tax into liquidity

The problem. Thai landowners who leave a plot unused pay a vacant-land tax that rises every three years, and selling costs about 3.3% in transfer tax. Investors, meanwhile, spend weeks paying people to check zoning before they buy. Most property sites list a price, but investors want to see the return.

Our solution. An AI land-use and tax simulator. The owner enters a title-deed number and the plot size, and the system:

Calculates the tax under current law and charts vacant against agricultural use
Analyses the location with GIS data to recommend the most valuable exit
Offers a one-click action: hire a partner to farm the plot (cutting the tax), or send a feasibility report to B2B investors

Business model. A 3% commission on deals, B2B SaaS subscriptions for developers and architects, and pay-per-use zoning feasibility reports. We asked for 200,000 THB in pre-seed funding.

What we built. A high-fidelity UI prototype covering the whole user journey, the rules of the Land and Building Tax Act turned into a data flow, and a system architecture ready for the MVP.

Pitch deck (PDF) Project poster