Michele Guastalla

AI voice dubbing and synthetic speech platform — ~20 languages, enterprise clients in fitness and gaming

• Work across the entire stack as one of 3–4 engineers: React and TypeScript front end, GraphQL APIs, Python around the ML voice pipeline, and deployment on AWS with additional workloads on Azure and GCP.
• Owned machine translation end to end: a bring-your-own-credentials integration for clients already using external providers, plus a self-hosted in-house model sold as a paid alternative, billed as metered credits.
• Built client-facing resource management tooling for allocating compute across processing jobs, handling projects that generate thousands of hours of output in a month; now a direct revenue stream.
• Integrated Paddle as merchant of record, opening a B2C sales channel alongside the existing enterprise business.
• Resolved concurrency and race-condition issues in the platform's update handling, eliminating a long-standing source of inconsistent state for users.

Four-month contract between semesters

• Designed and implemented new features in a large PHP, JavaScript and AWS codebase.
• Worked directly with clients and project managers to scope custom solutions, and fixed legacy-code defects that were impacting revenue.

Gap-year placement, then university summers, across the Dublin and Tokyo offices

• Ran localization testing on unreleased titles, handled multilingual customer support, and provided IT support to a 200-person office.

• GPA 3.8/4.0 — Coursework: Natural Language Processing, Web Technologies, Analysis of Algorithms

• GPA 3.75/4.0 — Summa Cum Laude

• Co-authored "Application of Large Language Models to DDoS Attack Detection", published in the SmartSP 2023 proceedings (Springer LNICST) with USC faculty and PhD students.
• Further unpublished work on LLM-driven automated spear phishing, and a BERT-based classifier detecting misleading news from headlines and article metadata.

• Built a navigation app prototype that reached the finals of the European GNSS My Galileo Solution competition, earning €15,000 in prize funding.

• Recreated the original Metroid with a new level and mechanics, prototyped at game jams, and published a title on itch.io with a small team.