profile picture

Thomas Hsu

About Me

My name is Thomas Hsu, and I am a backend-focused software engineer interested in building scalable APIs and AI-powered processing systems. I enjoy designing reliable backend architectures, optimizing systems for performance and cost, as well as integrating LLMs into production environments.

Experience

July 2025 - Present

Freelance Developer

Johnny Novo on Instagram and TikTok

  • Built and deployed a serverless React-based website, powered by a flexible schema-driven Firestore data model, serving ~200 daily visitors, featuring interactive components and an admin interface for managing and publishing ranked datasets
  • Reduced Firestore operation costs by 32% through implementing data denormalization and caching strategies
  • Developed and utilized a CI/CD automation pipeline using GitHub Actions to enable continuous testing, automated build validation, and rapid production deployments, eliminating manual release overhead and improving production reliability
TypeScript React Google Firebase GitHub Actions Vitest
Chicken Wars Website
July 2025 - August 2025

Software Engineering Intern

@ Consulta Immigration

  • Benchmarked OCR and LLM APIs (AWS Textract, OpenAI GPT-4o, etc.) for performance and cost, generating insights to identify the most suitable solutions for AI-assisted information extraction and document filtering
  • Built an automated pipeline to validate user inputs by integrating Google Gemini to extract unstructured information from client-uploaded documents stored in AWS S3, cutting manual review time for admins by 20%
  • Engineered a document filtering workflow powered by Google Gemini to identify and remove irrelevant pages from client uploads with over 93% accuracy validated using Jest unit testing, resulting in a 15% reduction in document processing time
  • Refactored sequential LLM document-processing workflows into concurrent async API calls, reducing processing time by 40%
TypeScript NextJS ExpressJS MongoDB AWS GCP Jest

Projects

NYC Subway Tracker API

  • Built a real-time NYC subway tracking API, transforming raw static and realtime GTFS feeds into usable RESTful endpoints
  • Implemented a Redis caching to memoize expensive database operations, reducing API response times by up to 80%
Python FastAPI PostgreSQL Redis Docker
NYC Subway
GitHub Link

Project Management Tool

  • Developed a collaborative project management platform with real-time Kanban board functionality using NestJS and React
  • Implemented JWT-based authentication, role-based access control, and WebSocket integration for live project updates
TypeScript NestJS React PostgreSQL Docker WebSockets
Kanban
GitHub Link

Codebase RAG Tool

  • Built a RAG-based tool that indexes any public GitHub repository and answers natural language questions about the codebase
  • Implemented AST-aware chunking, semantic search with FAISS, and context-aware answer generation using the Gemini API
Python FastAPI FAISS Docker
RAG
GitHub Link

Neural Network Handwriting Calculator

  • A program that uses a convolutional neural network trained with PyTorch to complete handwritten arithmetic by users
  • Engineered an OCR pipeline using OpenCV to segment and preprocess user inputs for accurate model prediction
Python PyTorch OpenCV Matplotlib Jupyter Notebook
Calculator
GitHub Link

Contact Me