Hello, I'm

Boyu Liu
Software Developer& AI Researcher& Product Engineer

Boyu Liu

Specializing in

Machine Learning

Math & CS @ UIUC · Incoming M.Eng. EECS @ UC Berkeley · Building intelligent systems at the intersection of ML research and engineering.

Scroll

About

Hi, I'm Boyu Liu — a Math & CS student at UIUC, incoming M.Eng. EECS at UC Berkeley. I build at the intersection of machine learning research and software engineering.

I'm genuinely excited by new technology, and one of my favorite things to do is turn ideas into real products — from the first line of code to something people can actually use. There's something uniquely satisfying about shipping something end-to-end.

My research interests includes Robotics that generalizes in the physical world, reinforcement learning, time series question answering, AI agents, and LLM routing. On the engineering side, I care about systems and infrastructures that are reliable and well-designed for fast iteration.

RoboticsReinforcement LearningTime Series QAAI AgentsLLM RoutingFull-Stack

Work Experience

Research Intern

Jul 2025 – Sept 2025

Machine Intelligence Lab @ Westlake University

Extended Reinforce (ICML 2024) with visual perception (R3M + proprioception) for policy learning on FurnitureBench robotic assembly tasks. Built an end-to-end offline RL training pipeline enabling 50K+ step runs on variable-length trajectory data.

PythonPyTorchReinforcement LearningComputer VisionW&B

Full-Stack Software Engineer

Oct 2024 – Dec 2024

National Center for Supercomputing Applications

Built EACUE, a mobile-first survey platform for accessibility evaluations using Next.js + Node.js + MongoDB, with a dynamic question-routing engine and Docker containerization.

Next.jsReactTypeScriptMongoDBDocker
Python
C++
JavaScript
TypeScript
Swift
Go
PyTorch
TensorFlow
NumPy
Pandas
LangChain
React
Next.js
Node.js
FastAPI
Spring Boot
Svelte
PostgreSQL
MongoDB
MySQL
SQLite
Docker
Git
GitHub
Vercel
Supabase
Postman
Python
C++
JavaScript
TypeScript
Swift
Go
PyTorch
TensorFlow
NumPy
Pandas
LangChain
React
Next.js
Node.js
FastAPI
Spring Boot
Svelte
PostgreSQL
MongoDB
MySQL
SQLite
Docker
Git
GitHub
Vercel
Supabase
Postman

Projects

PulseAI preview

PulseAI

AI Stakeholder Simulation Platform

Full-stack platform that simulates AI stakeholder reactions to validate product ideas. Multi-agent LLM pipeline generates diverse synthetic personas in parallel, reducing simulation latency by 90%. Streaming chat APIs for real-time conversations with AI personas persisted to PostgreSQL.

Next.jsLLMsPostgreSQLReadableStreamZod

Rhythm

Voice-First iOS Productivity App

Voice-first task management iOS app using SwiftUI and MVVM. Modular Python/FastAPI backend for agent orchestration with structured tool calling. Supabase + PostgreSQL with Row Level Security, JWT auth, and 100% offline functionality with background sync.

SwiftUIFastAPISupabasePostgreSQLSwift Concurrency

ISIS Distributed Transaction System

Distributed Systems

Distributed transaction processing system in Go across multiple cluster VMs using TCP peer communication. ISIS-style totally ordered multicast ensuring consistent global ordering across live nodes. Achieved 1ms p95 latency in 3-node and 3ms p95 / 10ms p99 at 8 nodes.

GoDistributed SystemsTCPISIS Multicast
Twitch+ preview

Twitch+

Personalized Twitch Recommender

Full-stack Spring Boot application for searching Twitch resources with personalized content-based recommendations. React + Ant Design frontend, MySQL on AWS RDS, RESTful APIs via OpenFeign, Spring Security for auth, and AWS App Runner deployment.

Spring BootReactMySQLAWSSpring Security
PathForms preview

PathForms

Interactive Math Visualization Game

Web game for visualizing mathematical path and form concepts, developed for the Illinois Mathematics Lab. Built with React, TypeScript, Next.js, and D3.js.

ReactTypeScriptNext.jsD3.js
ColorTaiko preview

ColorTaiko

Interactive Math Visualization Game

Web game that visualizes mathematical concepts through color and rhythm mechanics, developed for the Illinois Mathematics Lab. Built with React, TypeScript, Next.js, and D3.js.

ReactTypeScriptNext.jsD3.js
E3 Mini-Benchmark preview

E3 Mini-Benchmark

LLM Benchmarking Framework

Benchmarking framework for Small Language Models on the Efficiency-Energy-Effectiveness triad. LoRA fine-tuning reducing trainable parameters to 2.4% and GPU memory by 80%. Hardware-aware inference engine measuring TTFT, throughput, and energy-per-token.

PythonPyTorchLoRASuperGLUEMMLU

Publication

TSAQA: Time Series Analysis Question and Answering Benchmark

B. Liu, et al.

Under Review, ACL 2026

Time SeriesLLMsBenchmarkNLP

Comparative Analysis of Encoder-Only, Decoder-Only, and Encoder-Decoder Language Models

B. Liu

ICDSE 2024· DOI: 10.5220/0012829800004547

Language ModelsTransformerNLPBenchmark

Blog

Coming soon.

Contact

Open to research collaborations, full-time roles, and interesting conversations.