Yeo Tzun Kai

Software Engineering Intern — Amili Pte Ltd

July 2024 — November 2024, Singapore. A survey application proof of concept, Qualtrics surveys, and internal prototypes.

• Built a survey application proof of concept with React, SurveyJS, MongoDB, and backend data capture.
• Designed and deployed Qualtrics surveys for operational and customer feedback.
• Worked with teammates to troubleshoot issues, clarify requirements, and deliver practical internal prototypes.

Worked with: React, SurveyJS, MongoDB, Qualtrics.

Software Engineer — Amili Pte Ltd

December 2024 — Present, Singapore. Automation, AWS access workflows, production troubleshooting, and an internal retrieval-grounded assistant.

• Designed Python and shell automations that streamlined recurring support workflows and reduced manual effort.
• Managed AWS IAM access requests and permission configurations in line with internal security policies.
• Troubleshot production issues across EC2, S3, IAM, and CloudWatch, using SQL to investigate and correct data issues.
• Developed and deployed an internal RAG chatbot that retrieves PubMed research through vector search and produces grounded LLM responses with automated CI/CD.

Worked with: Python, AWS, SQL, CI/CD, LLMs.

Languages

• JavaScript / TypeScript
• Python
• SQL

Cloud & DevOps

• AWS
• Docker
• CI/CD

Tools & Frameworks

• SvelteKit & React
• API development
• Qualtrics

Research-grounded AI assistant

Python, LLMs, Vector search, CI/CD.

An internal assistant that ingests PubMed articles, retrieves relevant passages semantically, and generates answers grounded in source material.

Retrieval is vector-backed, so an answer can always be traced to the passages it came from.

Shipped behind automated CI/CD with evaluation runs on every change.

Dynamic survey platform

React, SurveyJS, MongoDB.

A proof-of-concept survey workflow with reusable front-end components, durable response capture, and a lightweight backend integration.

Question types are declared as data, so a new survey is configuration rather than code.

Responses land in MongoDB with a schema that survives question edits.