VP_
ORIGINWORKPROJECTSCONTACTBLOG

VANSH PATEL

I build software for nuclear construction sites.

INITIALIZE_CONTACTINITIALIZE_CONTACTINITIALIZE_CONTACTVIEW_PROJECTSVIEW_PROJECTSVIEW_PROJECTS

LEXINGTON, KY
38.0406° N — 84.5037° W

SCROLL
01ORIGIN

Most of my work is turning messy real-world data into tools people actually use.

Right now that means software at The Nuclear Company, where I build mapping tools, walkdown data models, and pipelines that keep construction data moving. It's detail-heavy work where small mistakes cost real money, which turns out to suit me fine.

Before this I was at the University of Kentucky, first as a research fellow mapping opioid overdose hotspots for the state, then building AI research tools at the campus computing center. Along the way I've made forecasting models for Papa Johns delivery windows, iOS apps for speech practice and habit tracking, and reinforcement learning environments.

B.S. COMPUTER SCIENCE · UNIVERSITY OF KENTUCKY · 2025 — MATH MINOR · AI CERTIFICATE · 4× DEAN'S LIST

TYPESCRIPT/PYTHON/SWIFT/PYTORCH/TENSORFLOW/XGBOOST/NEXT.JS/DOCKER
02THE_WORK

JUN 2025 — PRESENT

Software Engineer

The Nuclear Company

  • >Designed hierarchical mapping tool with dynamic zoom capabilities
  • >Engineered object-based data modeling system for walkdown applications
  • >Devised automated workflows for nuclear construction data ingestion

OCT 2024 — APR 2025

AI Solutions Developer

Center for Computational Sciences, University of Kentucky

  • >Engineered research platform using Next.js, TypeScript, and Tailwind CSS
  • >Developed automated transcription system 2x faster than existing solutions
  • >Enhanced data retrieval capabilities across 5 university departments by 30%

MAY 2024 — AUG 2024

AI Research Fellow

Center of Applied AI, University of Kentucky

  • >Led geospatial data visualization for RADOR-KY opioid overdose prediction
  • >Created automated hotspot detection system using ArcGIS Pro
  • >Engineered data pipeline reducing map generation time by 17%
03SELECTED_WORK
01

TOKENUINELY

Per-file semantic headers and a local embedding index for agentic code retrieval. Generates compact natural-language summaries for every file in a repo and serves them to Claude Code over MCP, replacing thousands of grep tokens with one semantic call.

> Built content-hashed incremental indexing for cheap re-runs on large repos

PYTHON • MCP • CLAUDE • EMBEDDINGS • SQLITE

02

BIG_TALK

iOS app for practicing conversational speech, with on-device transcription, filler-word detection, and a multi-signal scoring engine that grades fluency and substance in real time.

> Designed a progressive scoring pipeline that rewards engagement over length

SWIFT • SWIFTUI • WHISPER • SPEECH FRAMEWORK

03

PAPA_JOHNS_ORDERING_SYSTEM

Machine learning system for optimized order forecasting and time-slot reservations, improving store operations for carryout and delivery.

> 92% accuracy in predicting store capacity and delivery windows

ML FORECASTING • SARIMA • XGBOOST • LSTM

04

TRACE

Blockchain platform transforming credential verification. Universities issue verified records, students store and share securely, and employers verify instantly.

> Integrated smart contracts and blockchain for secure verification

TYPESCRIPT • NEXT.JS • ARWEAVE • SMART CONTRACTS

EVERYTHING_ELSE — 9 MORE

MONSTER_HUNTER_ARSWIFT · AUGMENTED REALITY↗HABIT_FORMEDSWIFT · SWIFTUI↗BANDITSPYTHON · REINFORCEMENT LEARNING↗SHIPPINGENVPYTHON · SIMULATION↗BLACKBOXSWIFT · IOS↗PROTEIN_ANALYSISPYTHON · BIOINFORMATICS↗HORSE_RACING_PREDICTORPYTHON · XGBOOST↗
CRAMBERRITYPESCRIPT · DOCKER↗
CARDSGO↗
04TRANSMIT

Get in touch.

vanshpatel1023@gmail.com
GITHUB ↗LINKEDIN ↗BLOG ↗

VANSH PATEL

LEXINGTON — --:--:-- ET

>....................0