Joshua Lum

Hello, I'm

Joshua Lum

CS + Business @ USC · Interested in Product, BizOps & Engineering

Download ResumeView Projects
jalum@usc.eduLinkedInGitHub

About

Background & expertise

I'm a Computer Science and Business Administration graduate from USC, currently based in NYC. I'm love learning, and it's brought me deeper into the field than I ever expected. Whether it's co-authoring research on ML interpretability or building AI-powered products as an 8VC Fellow, I'm always taking the hands-on approach to growth. Right now, I'm interested in working at roles around the intersection of Product, Business, and Engineering.

8VC Engineering FellowNeurIPS 2024 (CaLM)Magna Cum LaudeDean's List

Education

University of Southern California

Viterbi School of Engineering & Marshall School of Business

B.S. Computer Science and Business Administration · May 2026

  • Head Peer Academic Leader
  • Chinese American Student Association (President)

Experience

Where I've built and shipped

Campus

Series B · 8VC / Founder's Fund

Software Engineering Intern (8VC Engineering Fellow)

New York, NY · May 2026 – August 2026

  • 1 of 30 selected 8VC Engineering Fellows, building AI-powered features to enhance student learning.
  • Shipped a DM-native AI assistant ("Sam") across 4 services — Go backend, React frontend, Go WebSocket server, and Python/LangChain — enabling replies and proactive outreach in student DMs.
  • Improved AI response quality by updating model metadata on assignment info, adding file upload support, and implementing tracing through Datadog.
Sam AI assistant DM interface on Campus

Sam — Campus AI Buddy in student DMs

Uthana

Seed · A16Z portfolio

Product & Engineering Intern

Los Angeles, CA · May 2025 – August 2025

  • Launched a live generative AI animation demo at SIGGRAPH 2025, viewed by 2,000+ game developers and ML researchers.
  • Built an automated Python workflow that processed and ranked 10K+ motion prompts using frequency and embedding-based cosine distance to generate monthly top-150 MoCap batches.
  • Reported on 10+ companies and technologies in AI animation to shape product features and partnerships.

SIGGRAPH 2025 — live real-time generative animation demo

Tamagotchi Lab, USC

Viterbi School of Engineering

Undergraduate Student Researcher

Los Angeles, CA · May 2024 – May 2026

  • Conducted ML/NLP interpretability research on how LLMs encode syntactic and semantic information, with applications to fairness-critical decision systems.
  • Co-authored a NeurIPS 2024 (CaLM) paper on GPT-2 causal reasoning across 144 attention heads using activation patching on 30K+ sentence pairs.
  • Co-authored research analyzing fairness and bias in LLM-based kidney allocation using real OPTN data.

Projects

Things I've built

The Intern

Proactive AI Coding Agent

Watch Demo

An always-on AI agent in Slack that triages Linear tickets, writes code on feature branches, opens draft GitHub PRs, and demos on Replit — powered by a multi-agent architecture on the Claude Agent SDK.

ReplitPythonClaude Agent SDKSlack APILinearGitHub API
  • →Continuous preview deployments via GitHub Actions — every agent-opened PR auto-deploys to a live *.replit.app URL.
  • →Integrated Slack Socket Mode, Linear/GitHub MCP servers, and Perseus semantic code search for scoped subagent tool access.

3D Interactive AI Companion

Unity + OpenAI

Watch Demo

A 3D interactive AI companion in Unity with Blender-modeled animations, Microsoft Cognitive Services for speech, and OpenAI-powered dynamic conversation synced to character animations.

C#UnityBlenderOpenAI APIMicrosoft Cognitive Services
  • →Full speech recognition, synthesis, and lip-synced character animations.
  • →Dynamic conversational AI driving real-time 3D character behavior.

Research

Publications & papers

CaLM @ NeurIPS 2024

Causal Interventions on Causal Paths: Mapping GPT-2's Reasoning From Syntax to Semantics

Isabelle Lee, Joshua Lum, Ziyi Liu, Dani Yogatama

Characterized causal reasoning in GPT-2 small by analyzing cause-and-effect sentences, finding syntax localized in early layers and semantic sensitivity in later attention heads.

Read paper →

arXiv preprint

Evaluating Large Language Models for Fair and Reliable Organ Allocation

Brian Hyeongseok Kim, Hannah Murray, Isabelle Lee, Jason Byun, Joshua Lum, Dani Yogatama, Evi Micha

Systematic evaluation of LLMs on kidney allocation using real OPTN data, revealing task-dependent demographic biases and the need for rigorous fairness metrics before clinical deployment.

Read paper →

Skills

Languages

PythonPyTorchC++C#GoJavaJavaScriptTypeScript

Frameworks & Tools

ReactNode.jsUnitySQLGit

Domains

AI / MLNLPFull-StackAgent SystemsProduct

Interests

Music · Hiking · Pickleball · Spikeball · Snowboarding · Cycling · Gaming

Contact

Let's connect

jalum@usc.edu

(702) 793-5286

LinkedInGitHubResume