whoami

Douglas Le

Software engineer studying Computer Science and Mathematics of Finance & Risk Management at the University of Michigan. I like building things from the database up.

~/about

I'm a CS student at Michigan with a second major in Math of Finance & Risk Management, currently looking for new-grad software engineering roles. My coursework spans data structures and algorithms, database systems, web systems, and machine learning, and I've spent the past couple of months building Spun, a full-stack social app, end to end — backend, mobile client, and infrastructure. Most recently I built Drape, an AI agent that puts together outfits from your own wardrobe, to learn how to build and evaluate LLM applications.

~/projects/spun

Spun

A social rating & review app for music — built solo, end to end

A Letterboxd-style diary for music: rate and review albums and songs, build ranked lists, follow other listeners, and discover what's popular. I designed and built the full stack myself — a REST API over a normalized Postgres schema, a React Native mobile client shipping to the App Store and Play Store, and the infrastructure behind both.

TypeScript Node.js/Express PostgreSQL React Native (Expo) Docker

Built a 40+ endpoint REST API in Node.js, Express, and TypeScript over a normalized PostgreSQL schema (15+ tables), using Zod request validation and JWT/bcrypt authentication to enforce data integrity across a social graph of ratings, reviews, lists, and follows
Integrated the Spotify Web API for album, song, and artist metadata with a Postgres-backed caching layer that persists catalog data on first lookup, eliminating redundant external calls and keeping rating and review pages fast on repeat visits
Built the React Native (Expo) mobile client with a custom design system and optimistic UI updates for likes, follows, and ratings, then shipped signed builds via EAS Build
Implemented account deletion, blocking, and reporting flows required for Apple App Store and Google Play approval, meeting platform review requirements for apps with user-generated content

Try it live → View code

~/projects/drape

Drape

An AI agent that builds outfits from the clothes you actually own

Photograph your clothes and Claude catalogs each item; then ask for an outfit (“dinner, 45 degrees, smart casual”) and an agent checks the weather, searches your wardrobe, and picks pieces that fit the occasion. I built it to learn LLM engineering properly: tool use, vision with structured output, an agent loop, and evaluation, with code — not the prompt — enforcing what the model is allowed to return.

TypeScript Claude API Node.js/Express PostgreSQL React Docker Fly.io

Built a tool-using agent loop on the Anthropic SDK with four tools (wardrobe search, weather, rule check, submit) and an 8-call cap; submissions are validated in code, so an invented item ID is rejected and the agent corrects itself instead of returning clothes the user doesn't own
Extracted clothing attributes from photos with Claude vision and Zod-validated structured output, after resizing images and stripping EXIF/GPS metadata (~2¢ per photo), with a script measuring per-attribute accuracy against hand labels
Wrote an evaluation harness of 18 test requests × 3 runs, graded by deterministic checks against each request's expected temperature and formality; it caught a morning-vs-daytime forecast bug, and the fix raised the all-checks pass rate from 94% to 100% with no regressions
Made it multi-user and safe to run publicly: scrypt password hashing and hashed session tokens, per-user data isolation enforced at the agent's tool layer, quotas and a daily spend cap, and live agent progress over server-sent events that cancels the model call when the user leaves

Try the demo → View code

~/projects/other

Scalable Search Engine

Architected an end-to-end search engine as three decoupled services: a MapReduce indexing pipeline, a REST API index server, and a Flask search UI
Built a segmented, 134MB inverted index over 3,000+ documents, combining tf-idf and PageRank to rank by relevance and authority

Python Flask REST API TF-IDF/PageRank

Full-Stack Instagram Clone

Migrated a social app from Jinja2 server rendering to a React SPA backed by a 34-endpoint REST API, enabling live feed, like, and comment updates without page reloads
Deployed to AWS EC2 with Nginx as a reverse proxy and Gunicorn as the WSGI server

React Flask REST API AWS EC2 Nginx Gunicorn

Grace Hash Join

Implemented Grace Hash Join in C++ to join relations up to 999 pages against a 16-page memory buffer
Coordinated page-level I/O between emulated disk and memory buffers, mirroring how database engines join datasets larger than RAM

C++

Distributed MapReduce Framework

Built a distributed Manager/Worker execution framework inspired by Google's MapReduce paper, using TCP sockets to coordinate task distribution across independent Worker processes
Handled Worker failures by tracking heartbeats, marking Workers dead after a 10-second timeout, and requeuing their in-progress task for reassignment to the next available Worker

Python Distributed Systems TCP Sockets

~/skills

Languages

C++ Python Java JavaScript TypeScript SQL

Frameworks & libraries

Flask React React Native Node.js/Express Claude API NumPy Pandas

Tools

Git Linux PostgreSQL MongoDB Docker AWS (EC2) Fly.io Nginx Gunicorn

~/education

University of Michigan — Ann Arbor

Ann Arbor, MI

B.S. Computer Science and Mathematics of Finance & Risk Management · GPA 3.5/4.00 · University Honors

Coursework: Data Structures & Algorithms, Database Management Systems, Web Systems, Machine Learning, Bayesian ML

~/experience

Restaurant Manager

June 2024 – Aug 2025

Built and maintained the business's public website (HTML, CSS, JavaScript, jQuery, Bootstrap).