Kaung's SpaceSay hi
SWE · Product & Applied AI

Prologue: hello, I'm

Kaung Nay Lin Khant

Building backend services, browser tooling, and applied-AI products, one shipped chapter at a time.

start hereRead the storyConnect on LinkedIn
Chapter I

The Journey

From a self-directed CS degree to shipping production code across fintech, AI, and travel.

where it all started: studied while working

Jan 2022 – Apr 2025

B.Sc. Computer Science

University of the People · Remote

first real job: AR in the browser!

Jul 2023 – Jan 2025

Junior Frontend Engineer

rangoon.tech

  • Built a reusable WebAR SDK in TypeScript, Three.js, and React Three Fiber, integrating browser-based MediaPipe tracking without server-side computer vision.
  • The SDK was used by two engineers across 5+ enterprise WebAR projects for banking and insurance clients and supported major mobile and desktop browsers.
TypeScriptThree.jsReact Three FiberMediaPipe
the LLM era begins

Feb 2025 – Dec 2025

AI Engineer · EdOmnis LMS

ConceptX Company Ltd.

  • Delivered LMS-integrated OpenAI workflows and hosted an 8B DeepSeek-R1 distilled model on AWS EC2; implemented a FastAPI/SQS/DynamoDB queue so jobs could wait safely while the model worker was offline.
  • Built an Edexcel IGCSE English grading prototype from ~100 historical responses; achieved held-out mean absolute error of 1 mark for AO5 and 2 marks for AO4.
FastAPIAWSDynamoDBSQSLLMs
back to school, for on-device AI

Aug 2025 – Expected May 2027

M.Sc. Computer Engineering

King Mongkut's University of Technology Thonburi (KMUTT) · Bangkok, Thailand

  • Thesis (architectural design in progress): a four-phase edge/cloud memory architecture for an 8GB ARM device, evaluating peak RAM, latency, and retrieval quality.
zero → App Store in 5 months

Jan 2026 – May 2026

Software Engineer · Plutus Land

ConceptX Company Ltd.

  • Built the initial web and mobile product for a fractional-property platform using Next.js, React Native, and TypeScript, covering property investment, investor grouping, MMQR payment, and receipt-upload flows.
  • Released the React Native app to the App Store and Google Play; owned build configuration, store submissions, and initial AWS EC2/Cloudflare deployment while coordinating with two backend engineers and one designer.
Next.jsReact NativeTypeScriptAWSCloudflare
first big-tech deploy!

May 2026 – Jul 2026

Software Engineer Intern · Supply Extranet

Agoda Company Pte. Ltd.

  • Implemented and deployed an external travel-partner property-onboarding integration in C#/.NET, building the endpoint, service logic, and validation for importable property links.
  • Added experiment components, a monitoring dashboard, and error tracking for the production A/B rollout; diagnosed rollout failures and supported a controlled experiment restart before full evaluation.
C#.NETA/B TestingGrafana
every chapter led here

Aug 2026 – Present

Software Engineer Intern · Supply GenAI

Agoda Company Pte. Ltd.

  • Contributing to Agoda's Supply AI assistant, which helps Business Development get answers from supply data through chat, powered by MCP tool integrations.
MCPLLMsChat Tooling
Chapter II

The Craft

Personal and research builds. The work I shipped for companies lives back in Chapter I.

44% → 86%

valid-reasoning accuracy, English

Graduate NLP project · 2026

MF-Reason

English–Burmese Cultural-Reasoning LLM

A fine-tuned bilingual reasoning model built around a dataset designed from scratch for English–Burmese cultural context.

  • Implemented the complete pipeline for a 1,550-example English–Burmese cultural-reasoning dataset: source processing, LLM-assisted question and rationale generation, bilingual pairing, and human-reviewed preparation.
  • Fine-tuned Qwen2.5-3B-Instruct with 4-bit LoRA and Unsloth (1,240 train / 310 test); improved project-defined valid-reasoning accuracy from 44.52% to 86.45% in English and 1.29% to 68.39% in Burmese.
LoRAUnslothQwen2.5NLPBilingual
the one I'm proudest of
Pixel-art character generated by PixSprite: the base sprite next to its weapon-holding variant

prompt: "pixsprite, dragon"

Graduate Generative AI project · 2026

PixSprite

Paired Pixel-Sprite Generation Pipeline

A two-stage generative pipeline that turns base sprites into weapon-holding variants while keeping a consistent pixel-art identity.

  • Implemented a two-stage pipeline over 8,000 sprite pairs, fine-tuning Stable Diffusion 1.5 with LoRA and training a ControlNet conditioned on base sprites.
  • Developed deterministic mode-pooling from 32×32 to 16×16 and shared-palette K-means post-processing; evaluated LoRA-only, img2img, and ControlNet variants with identity and pair-accuracy metrics.
Stable DiffusionLoRAControlNetGenerative AI
Chapter III

The Toolkit

The languages, frameworks, and delivery habits I reach for most.

Languages

C#PythonTypeScriptJavaScriptSQL

Backend & APIs

.NET CoreFastAPIFastMCPNode.jsExpress.jsREST APIs

Cloud & Data

AWS (EC2, SQS, DynamoDB)PostgreSQLMongoDBDockerLinuxCloudflare

Product & Web

ReactNext.jsReact NativeThree.jsReact Three FiberMediaPipe

Delivery

A/B ExperimentsValidationError MonitoringGrafanaApp Store & Google Play Releases
Epilogue

Let's write the next chapter

I'm deep into product engineering and applied AI, and always down to build side projects, swap ideas, and collab with other builders.

Bangkok, Thailand

Best way to reach me is LinkedIn

Message me on LinkedIn

thanks for reading this far