# Thomas J. Vu

> Software engineer shipping agent systems, interactive products, and measurable AI infrastructure.

Based in Boston, Massachusetts.

- Email: hello@thomasjvu.com
- Resume: /thomas-vu-resume.pdf
- GitHub: https://github.com/thomasjvu
- LinkedIn: https://www.linkedin.com/in/thomasjvu
- Site: https://thomasjvu.com

## Selected work

- [LLM Let's Play - Fire Emblem](https://thomasjvu.com/projects/llm-plays-fire-emblem.md): Agent gameplay harness: mGBA state in, tactical context to the model, controller actions back, dashboard for humans.
- [SteezUI](https://thomasjvu.com/projects/steez-ui.md): Shadcn-style design system with dual install paths, inspectable demos, and craft-focused primitives for design-engineering portfolios.
- [Rally](https://thomasjvu.com/projects/rally.md): Waifu visual novel chat experience with Live2D, memory, and a relationship loop you can actually play.
- [Mystery Gift](https://thomasjvu.com/projects/mystery-gift.md): TEE-backed verifiable randomness infrastructure for giveaways and gachapon systems without on-chain costs.

## Research

- [BS-Bench](https://thomasjvu.com/writing/bs-bench-llm-deception.md): A 600-game benchmark for measuring LLM deception, lie detection, and instruction compliance. Honesty prompts reduced lying, but also reduced challenges and made remaining lies more successful.

## Writing

- [Lisper: AI Speech Therapy for Lisps](https://thomasjvu.com/writing/lisper-ai-speech-therapy-for-lisps.md): A quick writeup on Lisper, my Gemma 4 Good Hackathon submission for low-anxiety lisp speech practice.
- [BS-Bench: What a Bluffing Card Game Revealed About LLM Deception](https://thomasjvu.com/writing/bs-bench-llm-deception.md): A 600-game study of six language models playing Bullshit found that honesty prompts reduce deception, but also soften enforcement around the lies that remain.
- [Hardware Over Oracles: $0.01 TEE-Backed Randomness API (Offchain)](https://thomasjvu.com/writing/hardware-over-oracles-tee-randomness-api.md): Chainlink VRF is overkill for gacha games. We built a $0.01 alternative using Intel TDX.