Founded Ekkhara, a self-funded software & AI studio building EdTech products for Myanmar — taking each from idea to live product and owning architecture, backend, AI, and front-end end-to-end.

Built SpeakProof, a TOEFL speaking & English-practice bot that runs entirely inside Telegram — letting Myanmar learners train for the TOEFL computer-based test from an app they already use daily, with no VPN needed despite the country's internet restrictions.

Engineer the full stack behind each product — Python / FastAPI services, LLM-driven feedback and scoring, and conversational UX — shipping AI tools that reach users where access and infrastructure are constrained.

Architected and built an end-to-end Generative AI platform from zero — LLM orchestration, RAG pipelines, and a production React/TypeScript frontend with Python (FastAPI) and Azure Functions serverless backend

Designed and implemented responsible-AI validation layers including content-safety filters, bias-detection checks, and compliance guardrails ensuring all AI-generated outputs meet regulatory and ethical standards

Built AI safety and IP compliance systems including audit logging, provenance tracking, and output attribution pipelines — ensuring traceable and accountable AI output lifecycles

Developed prompt-engineering and LLM evaluation frameworks using Azure OpenAI GPT-4o and LangChain, with systematic fine-tuning to optimise domain accuracy across compliance use cases

Designed a cloud-based batch data processing pipeline using Microsoft Azure Batch and Docker for urban mobility analytics

Automated Computer Vision and Bluetooth sensor workflows enabling scalable, real-time smart city insights

Delivered production analytics dashboards adopted by city government and transport stakeholders

Engineered back-end services and data pipelines for a research lab delivering applied-AI capabilities to partner organisations — productionising computer-vision and NLP research behind Python APIs

Built a complete English–Myanmar machine translation system on a 10K gold-labelled WikiHow corpus — owning dataset construction, model training, evaluation, and deployment behind a public API

Developed novel research prototypes in computer vision and multilingual NLP, plus the back-end infrastructure to serve them — turning lab research into reproducible, deployable tools

Built an NLP paraphrasing tool end-to-end at a research lab providing applied-AI services to companies — from model development to back-end API

Engineered the data layer — cleaning and structuring training datasets — and optimised training workflows, improving model accuracy by 15%

Owned back-end operations for the lab's AI tooling, strengthening data-management and R&D engineering practice

Built full-stack web applications for a United Nations agency (FAO) — internal data dashboards and public-facing portals delivering agricultural and food-security programme data to staff and field teams

Engineered field data-collection tools and the back-end APIs behind them, letting field officers capture, validate, and sync programme data into centralised reporting systems

Owned features end-to-end across the stack — React/Next.js + TypeScript front-ends with Python and PHP/CMS back-ends — from data model and REST API to deployed UI