VINHTRAN

West Lafayette, Indiana

• Refactored the lab's public EMSuite cryo-EM server, used by over 2,400 users, to isolate each algorithm's code, cutting the work of integrating a new one.

• Held the refactor's behavior constant by replaying the old and new code paths against all 10,219 jobs in the production database and comparing their output.

• Resolved recurring web-server outages by moving static-file serving off the Python application to Apache, freeing the WSGI worker pool for requests that actually needed it.

• Migrated an untracked production service into Git, replacing direct edits on the server with reviewed pull requests.

• Added automated pytest coverage to a codebase that had none, which is what made the refactor safe to attempt at all.

• Ramped up on structural biology with PyMOL, comparing AlphaFold2 and AlphaFold3 predictions against experimental PDB structures.

• Python
• Django
• pytest
• Apache
• Linux
• Git
• PyMOL

Hanoi, Vietnam

• Scaled server-sent-event delivery to over 1,000 concurrent clients per instance by handling write backpressure, supporting three times the user growth.1,000+concurrent clients

  Led a team of five interns building a price service microservice that streams live stock prices over server-sent events.

• Price update delay reduced from 500 milliseconds to 80 milliseconds.500ms80msprice update delay

  Cut price-update delay by moving distribution onto Redis pub/sub, letting traders react to market movement sooner.

• Chart loading latency reduced from 165 milliseconds to 65 milliseconds.165ms65mschart load

  Cached candlestick data in Redis as fixed-size blocks under shared keys, so price charts load from memory rather than re-querying on every view.

• Page load time reduced from 300 milliseconds to 165 milliseconds.300ms165mspage load

  Optimized PostgreSQL queries by eliminating N+1 fetches and adding composite indexes.

• TypeScript
• TimescaleDB
• Redis
• Docker
• SSE

Hanoi, Vietnam

• Revamped an e-commerce backend made of order, payment, and inventory services on Spring Boot, deployed to GCP Cloud Run with rolling updates.

• Order processing time reduced from 800 milliseconds to 400 milliseconds.800ms400msorder processing

  Halved order processing time by replacing synchronous REST calls between services with Kafka-based event streaming.

• Built a change-data-capture pipeline with Debezium and Kafka to sync CockroachDB mutations into Elasticsearch, holding replication lag under a second.

• Average API response time reduced from 350 milliseconds to 85 milliseconds.350ms85msAPI response

  Introduced a Redis caching layer for product catalog and category endpoints.

• Developer environment setup reduced from two hours to ten minutes.2 hrs10 minenvironment setup

  Containerized every service with Docker, cutting local environment setup from two hours to ten minutes.

• Java
• Spring Boot
• Kafka
• CockroachDB
• Debezium
• Elasticsearch
• Docker
• GCP

Hanoi, Vietnam

• Developed a data pipeline with Web3.js that fetches and processes on-chain transactions, powering real-time hedging analytics.

• Query latency reduced from 320 milliseconds to 175 milliseconds.320ms175msquery latency

  Designed a PostgreSQL schema on AWS RDS to handle read and write traffic at scale.

• Built a React dashboard consuming GraphQL APIs, displaying queryable analytics with sub-second refresh from Redis.

• Instrumented six services with PromQL metrics and Grafana alerts on ingestion staleness, catching pipeline stalls that never raised an error.

• React
• Web3.js
• PostgreSQL
• AWS RDS
• GraphQL
• Grafana
• PromQL