syna
gogte

senior @ usc (cecs).
making hardware do cool, fast ai things.
twice at nvidia.
research with acme & anrg labs @ usc.
cecs outstanding academic achievement award.
minion lover.

let's chat: sgogte@usc.edu

in CECS

2×

nvidia

210+

students taught

github linkedin spotify resume (sw) resume (hw)

// where i've been

nvidia

santa clara
may–aug 2025

most recent

systems software engineering intern

profile-guided optimizations → 20% faster execution, millions in annual savings
built a runtime tracing tool that cut bug resolution time in half

nvidia

santa clara
may–aug 2024

systems software engineering intern

integrated bullseye coverage into gpu system processes (c), surfaced 20% total code coverage
implemented gsp-rm coverage collection — 10% fewer code bugs

usc viterbi

los angeles
aug 2023–now

course producer — ee109 / ee250 / ee345/ cs353 (half of the viterbi classes tbh)

teaching embedded systems + distributed systems/ networking + digital circuitsto 210+ students
weekly labs, office hours, grading — basically a ta but cooler + more fun + able to meet more peers

usc anrg

los angeles
aug 2024–now

undergraduate researcher

scheduling algorithms for conditional task graphs — targeting 20% makespan reduction

usc acme

los angeles
aug 2023–may 2024

undergraduate researcher

fine-tuned rnns for human activity recognition — improved accuracy of the system by 50%

bmc software

santa clara
may–nov 2023

r&d intern

built an internal chatbot with flask api + python

hacksc

los angeles
2023–now

community lead

organized workshops, speakers, events for usc's annual hackathon — 500+ people

// things i've built

llm · ml

fine-tuning llama 3.2

custom pytorch pipeline with lora + mixed precision. trainable params down to 0.113% of the model. 35% less memory, 2× faster — single 16gb gpu.

pytorchlorallamacuda

github ↗

ml · hardware

squeezenet on fpga

full squeezenet architecture for inference on an fpga board — quantization, all modules. trained in python to get weights, then deployed directly on hardware.

pythonsystemverilogquartusfpga

github ↗

hardware · fpga

connect 4 on fpga

rtl state machine in verilog. full connect 4 on real hardware, no soft cpu cheating.

verilogrtlfpga

github ↗

ml · hardware

cnn on fpga

trained in python, forward pass deployed in systemverilog for variable inputs. nios ii + c for backprop. ~80% accuracy on actual silicon.

systemverilognios iicmodelsim

github ↗

iot · audio

spotify song recommender

raspberry pi + grove sensors read ambient noise → picks from your spotify top tracks. 95% accuracy. yes, this was a class project.

pythonflaskraspberry pispotify api

github ↗

full stack

restaurant tracker

yelp api + sql backend + clean frontend. search restaurants, save favorites, book tables.

javasqlyelp apijs

github ↗

// the stuff that makes me not a robot

music

hyperpop when i'm energized, ambient when i'm in the zone, 2010 bangerz on road trips — non-negotiable. i have opinions about playlists and the shuffle algorithm is personally offensive to me.

→ see what i'm listening to

currently into

☕ hunting the best matcha latte in LA, follow me on beli @sgogte
📖 reading anything about how things get built
🏔️ griffith trails, runyon, eaton canyon
📌 maintaining 47 pinterest boards
🏋️ gym. a body building comp perhaps...
🍌 the minions!! and the new movie coming out this july
✈️ most importantly: senior year maxxing - reach out for my flighty!! (20 flights this year)