Skip to content

Daniel Fan

Data Scientist, Engineer,
Martial Artist, Magician

Download Resume
Daniel Fan

Intro

What I am all about.

I'm Daniel Fan, an Applied Scientist and Senior Machine Learning Engineer with 10+ years of experience building AI systems that serve tens of thousands of users worldwide.

My expertise spans transformers, LLM fine-tuning, reinforcement learning, and NLP. I've trained and deployed custom large language models, published peer-reviewed research at IJCNN, and led teams to deliver both scientific innovation and production impact.

I hold a Master's degree in Electrical and Computer Engineering from Binghamton University. Outside of tech, I'm also a martial artist and magician.

Skills

Technical overview.

I'm passionate about technology, so I'm always excited to learn something new.

These are the things that I'm the most confident in, and the ones that I use the most professionally.

LLMs & Transformers (GPT, T5, BERT, RoBERTa) 95%
Python & PyTorch 95%
RLHF & Fine-tuning (LoRA/PEFT) 90%
AWS (SageMaker, EC2, Lambda) 85%
NLP & Deep Learning 90%
MLOps & Model Deployment 85%

Experience

Batman would be jealous.

Machine Learning

Training, fine-tuning, and deploying LLMs with LoRA/PEFT and quantization to serve 30,000+ users worldwide.

Data Science

Building predictive analytics and anomaly detection systems that saved $800K annually and reduced downtime by 20%.

Research

Published peer-reviewed papers at IJCNN on neural networks and transformer architectures for NLP tasks.

Work History

Where I've worked.

Senior Machine Learning Engineer

Swayable | Seattle, WA | 2022 - Present

Built and scaled ML training systems running thousands of models weekly. Designed LLM-powered moderation that cut costs by 50%.

Head of AI / Senior ML Engineer

Advanced Learning Ltd. (Writer's Toolbox) | Auckland, NZ | 2020 - 2025

Led transformer research and deployed 10+ custom LLMs serving 30,000+ users. Secured $500K in R&D funding.

Data Scientist

Fonterra | Auckland, NZ | 2018 - 2020

Built predictive analytics saving $800K annually. Implemented anomaly detection reducing downtime by 20%.

Machine Learning Engineer

Debatebot | Remote | 2017 - 2019

Processed 60M+ academic papers into Neo4j knowledge graph. Built research infrastructure for transformer experiments.

Publications

Research & papers.

Fan, H.-T.; Wang, W.; Jin, Z. (2017). Performance Optimization of Echo State Networks through Principal Neuron Reinforcement. IJCNN 2017, pp. 1717-1723.

Wang, W.; Fan, H.-T.; Jin, Z. (2017). Structure Optimization of Dynamic Reservoir Ensemble using Genetic Algorithm. IJCNN 2017, pp. 2193-2200.

Fan, D., Hunter, I., & Liu, B. (2024). Model and Algorithm Development for Paragraph Quality and Logic Assessment in Student Writing. Stage 2 Research Report, Advanced Learning Ltd.

Fan, H.-T. (2017). Evolution of Echo State Networks with Anti-Oja Plasticity Rules. M.S. Thesis, Binghamton University.

Education

Where I studied.

M.S. Electrical and Computer Engineering

Binghamton University (SUNY) | Binghamton, NY | 2017

Thesis: Evolution of Echo State Networks with Anti-Oja Plasticity Rules. Research focus on neural network optimization and reservoir computing.

Portfolio

Things I've built.

LLM Moderation System
Hide

LLM-Powered Moderation System

Production ML System at Swayable

Designed and deployed an LLM-powered content moderation system that outperformed human benchmarks while cutting moderation costs by 50%.

Built and scaled a training system running thousands of ML models weekly, accelerating research velocity and deployment across the organization.

  • LLMs,
  • PyTorch,
  • AWS

Role: Senior ML Engineer

Company: Swayable

Impact: 50% cost reduction

Educational LLMs
Hide

Educational LLM Platform

Custom LLMs for 30,000+ Users

Led applied research on transformer architectures (GPT-2, T5, BERT, RoBERTa) for complex educational NLP tasks at Writer's Toolbox.

Trained, fine-tuned, and quantized 10+ custom LLMs serving 30,000+ concurrent users worldwide. Applied LoRA/PEFT and quantization for efficient deployment.

  • Transformers,
  • LoRA/PEFT,
  • NLP

Role: Head of AI

Company: Writer's Toolbox

Users: 30,000+

Grammar Error Detection
Hide

Grammar Error Detection

T5 Transformer Model

Trained a T5 model achieving 87% accuracy on multi-objective grammar error detection for noisy learner-generated text.

Developed a multi-label classification model (BERT) achieving 98% accuracy across an educational taxonomy.

  • T5,
  • BERT,
  • NLP

Role: ML Engineer

Accuracy: 87-98%

Type: Research

Knowledge Graph
Hide

Academic Knowledge Graph

60M+ Papers in Neo4j

Ingested and processed 60M+ academic papers into Neo4j, enabling large-scale knowledge graph construction for Debatebot.

Built research infrastructure for experimentation with transformer-based models in applied argumentation tasks.

  • Neo4j,
  • NLP,
  • Data Engineering

Role: ML Engineer

Company: Debatebot

Scale: 60M+ papers

Predictive Analytics
Hide

Predictive Analytics Platform

$800K Annual Savings

Built predictive analytics models at Fonterra that reduced transport costs by $800K annually.

Implemented anomaly detection systems that cut manufacturing downtime by 20%, improving operational efficiency across the supply chain.

  • Predictive Analytics,
  • Anomaly Detection,
  • Python

Role: Data Scientist

Company: Fonterra

Impact: $800K saved

Contacts

Contact me.

Top