BillQuach

BillQuach

Software Engineer, Full-stack & Applied ML, CS @ UCF

Claude Monet, Woman with a Parasol — Madame Monet and Her Son (1875): a woman in white on a grassy rise, holding a parasol against a bright, wind-blown sky.

~ $ whoami

bill quach — software engineer, cs @ ucf ’28

~ $ cat highlights.txt

+38.4%
throughput from KV-cache prefix cachingA/B harness · CUDA/PyTorch
1st ×2
Waymo (of 77) & State Farm (of 34) tracksShellHacks 2026 · Halo
<50 ms
vehicle detection on a Raspberry Pi 5YOLO via NCNN · Halo
17%
throughput loss traced to 121 over-capacity admissionsThunderAgent, ICML 2026
7 flags
silently ignored for 4 months, fixed & sent upstreamFastAPI · pytest

~ $

Claude Monet · Woman with a Parasol · 1875 · NGA (CC0)

About

Maker at honest scale.

I'm a CS undergraduate at UCF (B.S. expected May 2028) who builds full-stack software and applied ML — currently benchmarking LLM serving as an undergraduate researcher, and testing workshop code for KnightHacks. I like systems you can see the whole of: an ordinary figure working a large machine, and the machine drawn honestly.

  • Now Undergraduate Researcher · LLM serving
  • Also QA Engineer · KnightHacks
  • Focus ML systems · Computer vision · Full-stack
  • Open to SWE internships
Education

B.S. Computer Science · University of Central Florida

Expected May 2028 · Data Structures & Algorithms I & II · OOP · Discrete Structures · Algos for ML · Math for ML

Skills

Languages Python · C++ · C · Java · JavaScript · TypeScript · SQL

ML & Computer Vision PyTorch · TensorFlow · CUDA · vLLM · OpenCV · YOLO · NCNN · Hugging Face Transformers · CLIP · LangChain · ChromaDB · pandas · NumPy

Frameworks & APIs React · Node.js · Flask · FastAPI · Three.js · REST APIs · OAuth 2.0 · Gemini & Claude APIs

Tools & Hardware Git · GitHub · Docker · Linux · pytest · FFmpeg · SQLite · Jupyter · LaTeX · Raspberry Pi · Arduino

The Workshop

Selected Work

#01 ShellHacks 2026 winner

Halo

A blind-spot detection bike helmet that won two tracks at ShellHacks 2026.

1st / 77 Waymo track, ShellHacks 2026

Python · YOLO · OpenCV · NCNN · +4

#02 Full-stack · Applied ML

Pinlist

Turns Pinterest mood boards into playlists.

140+ images classified per call

Python · JavaScript · CLIP · PyTorch · +4

#03 Applied ML · NLP

PageSense

Semantic book recommendations over 7,000 titles.

7,000 titles searchable by meaning

Python · Hugging Face Transformers · LangChain · ChromaDB · +2

All projects
~/projects/halo✕

~/projects $ cat halo/README.md

ShellHacks 2026 winner

Halo

A blind-spot detection bike helmet that won two tracks at ShellHacks 2026.

A bike helmet that watches the rider's blind spots: YOLO vehicle detection on a Raspberry Pi 5, ultrasonic sensor fusion for distance, time-to-contact alerts over serial to an Arduino, and a live 3D threat view in the browser.

1st / 77
Waymo track, ShellHacks 2026
1st / 34
State Farm track
<50 ms
vehicle detection on a Raspberry Pi 5
15 Hz
3D threat view over Server-Sent Events

$ cat halo/NOTES

  • Won 1st of 77 projects for Waymo and 1st of 34 for State Farm at ShellHacks 2026 with a blind-spot detection bike helmet, leading its computer vision, sensor fusion, and real-time visualization.
  • Detected vehicles at sub-50 ms latency on a Raspberry Pi 5 with YOLO via NCNN, streaming time-to-contact alerts to an Arduino subsystem over non-blocking serial with a 1.5s watchdog failsafe.
  • Cut side-distance error from ±30% in meters to centimeter-level accuracy via sensor fusion with median-filtered ultrasonic sensors, citing verifiable information for Gemini API incident reports on 10s clips.
  • Built a real-time 3D threat visualization model in Three.js, rendering detected vehicles with threat severity and direction from 700-byte JSON snapshots streamed over Server-Sent Events at 15 Hz.

$ ls halo/stack

  • Python
  • YOLO
  • OpenCV
  • NCNN
  • Three.js
  • Gemini API
  • Raspberry Pi 5
  • Arduino

devpost ↗

~/projects $

← simplysolar1/4pinlist →
plate-01.jpg

Claude Monet · Woman with a Parasol · 1875 · NGA (CC0)

~/projects/pinlist✕

~/projects $ cat pinlist/README.md

Full-stack · Applied ML

Pinlist

Turns Pinterest mood boards into playlists.

A full-stack multimodal recommender that reads the mood of a Pinterest board and returns a playlist to match it — image understanding on one end, the Spotify catalogue on the other.

140+
images classified per call
20+
semantic mood labels
≥80%
similarity for top songs

$ cat pinlist/NOTES

  • Engineered a full-stack multimodal music recommendation app, generating personalized playlists from image boards by classifying 140+ images per call across 20+ semantic labels via CLIP zero-shot classification.
  • Achieved ≥80% similarity score for top songs across mood categories by embedding 50 candidate tracks into a shared text vector space and ranking via cosine similarity against pre-cached mood vectors.
  • Built a Flask backend integrating Pinterest and Spotify REST APIs for board and playlist retrieval via OAuth 2.0 with persisted session authentication in SQLite.

$ ls pinlist/stack

  • Python
  • JavaScript
  • CLIP
  • PyTorch
  • REST APIs
  • OAuth 2.0
  • Flask
  • SQLite

~/projects $

← halo2/4pagesense →
plate-02.jpg

Claude Monet · Woman with a Parasol · 1875 · NGA (CC0)

~/projects/pagesense✕

~/projects $ cat pagesense/README.md

Applied ML · NLP

PageSense

Semantic book recommendations over 7,000 titles.

A natural-language book recommender that searches by meaning, not keywords — combining vector similarity, zero-shot genre classification, and a per-title emotion fingerprint over a corpus of 7,000 books.

7,000
titles searchable by meaning
7
emotion dimensions per title
≥25
word floor for clean descriptions

$ cat pagesense/NOTES

  • Built a semantic book recommendation engine over 7,000 titles using vector similarity search with LangChain, ChromaDB, and OpenAI Embeddings, plus zero-shot genre and emotion classification.
  • Scored all book descriptions across 7 emotion embeddings (joy, fear, sadness, anger, disgust, surprise, neutral) via sentence-level inference and max-pooling aggregation to build an emotion fingerprint per title.
  • Engineered a multi-stage NLP data pipeline, performing missing-value imputation, Spearman correlation analysis, and description filtering (≥25 words) to produce a clean, ISBN-tagged corpus for downstream modeling.

$ ls pagesense/stack

  • Python
  • Hugging Face Transformers
  • LangChain
  • ChromaDB
  • OpenAI Embeddings
  • Pandas

~/projects $

← pinlist3/4simplysolar →
plate-03.jpg

Claude Monet · Woman with a Parasol · 1875 · NGA (CC0)

~/projects/simplysolar✕

~/projects $ cat simplysolar/README.md

Hackathon · BloomHacks · 12 hours

SimplySolar

A 3D solar-panel planner, built in 12 hours at BloomHacks.

Type in an address and get an interactive 3D model of the roof with solar panels you can place, scored for placement and price by a parallelized Claude/Gemini reasoning pipeline. Led a 4-person team.

~50%
latency cut via parallelized AI pipeline
12 hrs
hackathon build, 4-person team
4
people led

$ cat simplysolar/NOTES

  • Led a 4-person team to build a 3D solar-panel planner in 12 hours at BloomHacks — type in an address and get an interactive 3D model of the roof with panels you can place.
  • Scored panel placement and price with a parallelized Claude/Gemini reasoning pipeline, cutting latency by ~50%.

$ ls simplysolar/stack

  • React
  • TypeScript
  • Three.js
  • Node/Express
  • Flask
  • Claude API
  • Gemini API
  • Google Solar API

~/projects $

← pagesense4/4halo →
plate-04.jpg

Claude Monet · Woman with a Parasol · 1875 · NGA (CC0)

~/experience/undergrad-researcher✕

~/experience $ cat undergrad-researcher.md

Aug 2026 — present

Undergraduate Researcher

College of Engineering and Computer Science, UCF

$ cat undergrad-researcher/NOTES

  • Built a scripted A/B benchmark harness on CUDA/PyTorch with rotated run order and per-run cache instrumentation, measuring 38.4% higher throughput from KV-cache prefix caching.
  • Traced a 17% throughput loss in ThunderAgent (ICML 2026) to 121 over-capacity admissions using the same harness, cutting cache hit rate 0.82 to 0.65 and 5x’ing p90 latency to 16.6s.
  • Fixed a FastAPI proxy defect that silently ignored 7 CLI flags for 4 months, self-looping every request and blocking all in-repo reproduction scripts; submitted upstream with pytest regression tests.

resume.pdf ↗

~/experience $

← qa-engineer1/2qa-engineer →

Claude Monet · Woman with a Parasol · 1875 · NGA (CC0)

~/experience/qa-engineer✕

~/experience $ cat qa-engineer.md

Jan 2026 — present

Quality Assurance Engineer

KnightHacks

$ cat qa-engineer/NOTES

  • Debugged and integration tested demo code and repos across ML, database, and front-end stacks for 10+ technical workshops delivered to 250+ attendees while coordinating across a 15-member team.
  • Promoted workshop attendance and social media presence by collaborating with a 30+ member cross-functional Outreach and Design team.

resume.pdf ↗

~/experience $

← undergrad-researcher2/2undergrad-researcher →

Claude Monet · Woman with a Parasol · 1875 · NGA (CC0)