Computer VisionResearch
ChessVision
Reached 90%+ move detection accuracy for real-time over-the-board chess digitization.
PythonRoboflowOpenCVYOLOv11
Duke Math + CS ’28 · Software & ML
James Wright
Saynario (HackPrinceton) · MINGL bioRxiv co-author · Code+ SWE on STINGAR
Durham, NCSeeking Summer 2027 SWE / quant internships · US citizen, authorized to work in the US
I build software at the edge of machine learning, data, and systems.
Duke Mathematics + Computer Science · Software engineer & ML researcher
Chess-trained intuition. Engineering-first execution.
USCF Candidate Master · Peak top 100 Lichess rapid (2538)
Voice products, bioRxiv spatial proteomics, and 15M-row health data pipelines.
Four concurrent campus roles · Code+, DIIG, Duke AML, Hickey Lab
2×
Hackathon winsHackathon wins · ElevenLabs (HackPrinceton), Solana (HackDuke)
1,000+
ICD-10 · 15M+ rowsICD-10 codes mapped · Gradient Health dataset (15M+ rows)
bioRxiv
MINGL preprintNamed co-author preprint · MINGL
4.0
GPA · Math + CSGPA · Mathematics + CS, Duke
~1500
Chess engine EloElo · self-built chess engine (Python & C++)
Top 100
Peak Lichess · USCF CMPeak Lichess rapid rank worldwide (2538) · USCF Candidate Master
More signal
99.5%
BERT · WELFakeBERT vs. classical embeddings · WELFake (known-separable benchmark)
4,000+
Discord NLPDiscord messages modeled · HayhaBots growth analysis
90%+
ChessVision CVReal-time chess move-detection accuracy
Top 0.04%
Chess.com top 0.04%Chess.com blitz · global percentile
[01] Projects
Selected work
selected project deck
Winner · live voice productVoice product
Saynario
Won HackPrinceton Best Use of ElevenLabs with a shipped voice roleplay loop.
More projects
SustainabilitySystems
CO2Ldown
Built a browser-extension and web workflow for per-action carbon estimates across sessions.
React.jsPythonHTMLCSS
NLP classifier
Fake News Detection
BERT vs. classical · WELFake
Research
Fake News Detection
Compared classical embeddings against BERT fine-tuning on WELFake — 99.5% test accuracy on a known-separable benchmark.
PythonBERTscikit-learnGensim
Chess engine
Chess Engine Development
~1500 Elo · C++/Python
Chess EngineSystems
Chess Engine Development
Deployed self-built Python and C++ engines that reached an estimated 1500 Elo on Lichess.
PythonC++MinimaxAlpha-Beta Pruning
[02] Experience
Where I've worked
Software Engineer · Duke University, Code+
May 2026 – Present · Full-time summerSummer software engineering on STINGAR, building LLM-driven honeypot prototyping and threat analysis tooling. STINGAR is an AI-powered cyberdefense platform used across 70+ partner universities.
- LLM-driven honeypot prototyping
- Threat analysis tooling for STINGAR
- Cyberdefense engineering on a university-scale platform
Data Analyst · Duke Impact Investment Group
September 2025 – Present · ~8 hrs/wk during termBuilt an AI-assisted ICD-10 coding workflow on Gradient Health’s 15M+ row parquet patient dataset through a DIIG partnership, and ran a dual-model NLP pipeline over 4,000+ Discord messages for case-discovery and growth insights.
- Mapped 1,000+ ICD-10 codes to de-identified cases for Gradient Health using an LLM pipeline on their parquet dataset
- Applied RoBERTa and VADER to a 4,000+ message Discord dataset for HayhaBots
ML Engineer · Duke Applied Machine Learning
September 2025 – Present · ~6 hrs/wk during termBuilt a longitudinal Alzheimer’s disease forecasting pipeline to predict next-visit ADAS13 scores using 3D MRI trajectory features and Neural Controlled Differential Equations.
- Extracted 3D MRI trajectory features across visits
- Modeled irregular time-series histories with Neural Controlled Differential Equations
Research Intern · Hickey Lab
September 2025 – Present · ~6 hrs/wk during termPublished a scverse-compatible Python package (MINGL) for probabilistic cell-type classification with 13 tool and 13 plotting functions spanning gradient, border, and heterogeneity analyses.
- Implemented Gaussian Mixture Models for cell-type classification
- Named co-author on a bioRxiv preprint
Four concurrent roles during the academic year — scope notes reflect typical weekly commitment alongside coursework.
[03] Quant / ML
Quant-relevant work
Research-grade modeling on irregular longitudinal data, DIIG investing-group pipelines on 15M+ row health data, and competitive search-and-pruning depth — maps to quant-adjacent SWE and ML engineering roles.
Neural CDE disease forecasting
Irregular longitudinal ADNI data with trajectory-only Val MAE ~13–16 and tabular MMSE ablations down to ~7.
Prompt-to-dataset financial pipelines
HackDuke Best Use of Solana — multi-agent orchestration with schema validation subagents for structured financial data from plain-English requests.
Probabilistic spatial proteomics
GMM-based cell-type classification packaged for scverse workflows; bioRxiv co-author.
Search, pruning, and competitive play
Self-built chess engines (~1500 Elo), USCF Candidate Master, and peak top 100 Lichess rapid (2538).
DIIG health-data & NLP pipelines
AI-assisted ICD-10 mapping on Gradient Health’s 15M+ row dataset and dual-model Discord NLP for case discovery.
Coursework: Combinatorics (MATH 371) · High Dimensional Data Analysis (MATH 465) · Probability (MATH 230) · Linear Algebra (MATH 221) · Advanced Multivariable Calculus (MATH 222)
Resume
PDF · Last updated June 2026
[04] Skills
Technical stack
Languages
PythonTypeScriptJavaC++C
ML & Data
PyTorchtorchcdepandasNumPyScikit-learnGaussian Mixture ModelsHugging FaceOpenCVRoboflow
Systems & Web
ReactNext.jsFastAPIWebSocketsLangGraphFirebaseParquetDigitalOcean
AI & Tooling
Gemini APIWhisperElevenLabsNLTKMatplotlibKerasCrawl4AI
[05] Background
Education
Duke University
Mathematics and Computer Science · GPA 4.0
Durham, NC · August 2025 – May 2028
Coursework
- Combinatorics (MATH 371)
- Introduction to High Dimensional Data Analysis (MATH 465)
- Probability (MATH 230)
- Advanced Multivariable Calculus (MATH 222)
- Linear Algebra (MATH 221)
- Data Structures and Algorithms (COMPSCI 201)
- Intro to Computer Systems (COMPSCI 210)
Thomas Jefferson High School for Science and Technology
GPA 4.576/4.0 weighted
Alexandria, VA · August 2021 – June 2025
Coursework
- Linear Algebra
- Multivariable Calculus
- Artificial Intelligence I & II
- Machine Learning I & II
[06] Recognition
Achievements
MINGL preprint
Named co-author on a bioRxiv preprint through the Hickey Lab.
USCF Candidate Master
Peaked at #41 nationally for age 18 (June 2025). Three-time National Team Champion and two-time National Team Runner-Up.
Online chess rankings
Peaked at rank 100 in rapid worldwide on Lichess.org (rating 2538) and top 0.04% of Chess.com blitz players.
Virginia College State Champion
Won the Virginia College State Championship in March 2026. Six-time Virginia state champion across team and individual events.
Scholastic & open chess
Won the Cherry Blossom Classic U2000 with a perfect 9/9 score. Featured in The Washington Times for a Round 1 upset over an International Master at the 2023 North American Junior U20.
[07] Contact