About Experience Work Publications Contact

UCSB · Undergrad Researcher

Dylan Lu.

I research AI agents, long-horizon reasoning, multi-agent collaboration, and reinforcement learning.

AffiliationUC Santa Barbara

ProgramCCS Computer Science

FocusNLP, RL, Agents

Scroll to explore

About

Things I
work on.

Undergraduate at UC Santa Barbara (College of Creative Studies), working with the UCSB NLP Group under Xin Eric Wang. I work on AI agents, long-horizon planning, and multi-agent collaboration.

I also won the grand prize at Cal Hacks, the world’s largest collegiate hackathon.

1st Place

Grand Prize · Cal Hacks

4.0

GPA

Experience

Research and
industry.

Summer 2026

ML Research Intern

Simular · Autonomous Computer Agents

Building AI agents that autonomously operate computers and automate workflows across macOS, Windows, and Linux.

Incoming

Spring 2026

ML Research Intern

ChipAgents · Agentic AI for Chip Design

Agentic AI platform automating semiconductor design and verification workflows.

Incoming

Oct 2025 —

Undergraduate Researcher — NLP

UCSB NLP Group · Advisor: Xin Eric Wang

Research in multimodal and agentic AI — embodied agents, language grounding, and reinforcement learning.

Research

Jun — Sep 2025

ML Engineer Intern

TetraMem · AI Chip Startup

Optimized Rust-based ML compiler: 16x faster depthwise convolutions, reduced compile time from 40 min to 2 sec. Built CI/CD pipeline and custom Halide C++ autoscheduler.

Industry

May — Aug 2022

Software Engineering Intern

Atmosic Technologies · IoT Chip Startup

Developed C-based QA pipeline accelerating IoT testing 3×. Built Python benchmarking tools for energy efficiency profiling.

Industry

Selected Work

Projects and
hackathons.

★ Grand Prize — Cal Hacks 12.0

FaceTimeOS

Computer-use AI agent enabling remote Mac control via FaceTime and iMessage with real-time speech interaction. Won 1st place among 695 projects at the world’s largest collegiate hackathon (3,000+ participants).

FlaskReactElectronSpeech AI

Reinforcement Learning

AI Agents for Geometry Dash

Mixture-of-Experts DQN agent achieving 90% human-level completion with a real-time game interaction pipeline.

MoE DQNOpenAI GymC++

Voice AI

RAG Voice AI Agents

On-device voice agent with <1 s speech latency and +12 point answer quality improvement via GraphRAG retrieval.

Voice LLMLangChainGraphRAG

★ 1st Place Entertainment — SBHacks

Baddy Buddy AI Coach

Badminton coaching system with 94% tracking precision using ViT models, deployed to 15+ athletes for real-time strategy insights.