CS @ USC | Incoming SWE Intern @ NVIDIA | Published Researcher | 4x Hackathon Winner

Sophomore studying Computer Science at USC Viterbi. Incoming Software Engineering Intern at NVIDIA building agentic AI. Published researcher at ICWSM 2025. Selected for selective quant discovery programs such as Jane Street FOCUS and Hudson River Trading – Inside HRT.

First-generation Mexican-American from Azusa, California.

I also bench 225 now. and still play soccer...

Voice-controlled Claude agent that operates a real iPhone — built for blind and low-vision users. Hold the Action Button, speak a task, watch Claude take over any app (LinkedIn, CVS, your bank) while AVSpeechSynthesizer narrates every step. 20+ tools, CoALA-style memory stack, prompt caching, 10–100× action-latency reduction via direct HTTP to Maestro's XCTest runner.

🥉 3rd Place — Claude Builder Hackathon @ UCLA, April 2026

An immersive 3D portfolio built as an interactive auto body garage. Visitors explore my work by clicking on garage objects — smooth camera fly-ins, GPU-tier quality scaling, and demand-driven rendering (0fps when idle).

Live at bryanram.com

Award-winning agentic AI platform for cost segregation. Multi-agent vision pipeline (Grounding DINO → SAM2 → GPT) with IRS-grounded RAG system. 2 paying customers from top-5 firms.

Won LavaLab Demo Night F'25 (Best Traction)

Quantitative fantasy sports analytics. Poisson modeling, 100k-run Monte Carlo simulation, GARCH(1,1) volatility forecasting, and cross-platform arbitrage engine.

78% win rate — turned $10 into $3,000

Agentic procurement co-pilot built at Caltech's HackTech. Event-driven LangChain agent with intent classification and multi-step reasoning. Invited by Dryft to SF Neo offices for post-hackathon collaboration.

1st Place Dryft Challenge @ Caltech HackTech

iOS-Agent: Open Research Platform for Physical Mobile GUI Agents — USC ISI · Jan 2026 – Present

First open research platform for autonomous AI agents on physical iOS. 22+ tool primitives, first autonomous multi-step drag on iPhone, 3.5× speedup on 15-step agent tasks (88s → 24s) via context compression, direct XCTest HTTP, and prompt caching.

EDTok: A Multimodal Eating Disorder TikTok Dataset — Accepted to ICWSM 2025 · arXiv:2505.02250

LLM-assisted pipeline for building the first multimodal eating disorder dataset from TikTok content.

USC ISI · Research Intern iOS-Agent (current) · EDTok @ ICWSM 2025	Amazon × USC · Research Intern LLM Safety (Jan 2026 – Present)	USC Viterbi · CS Sophomore

Full list on bryanram.com.

12 Hackathons. 4 Wins. Coast to Coast.

Languages

AI/ML

Web

Backend & Cloud

🌐 bryanram.com · 📄 Resume · 📧 bryanram2024@gmail.com · 💼 LinkedIn