方威 — the author's name in Chinese calligraphy

Ethan Fang

Based out of the Bay Area. I spend most of my time in math modeling, AI research, and engineering.

I graduated from Case Western Reserve University in May of 2026 with a BS/MS in computer science. During my time there, I pursued many interests such as: starting the quant finance club, growing club tennis, and doing lots and lots of research.

I enjoy building cool and useful things. This includes a startup I cofounded, Vantura, and a few sites such as iLiterate.

Currently, I am a quantitative analyst at KeyBank where I model expected loss for the bank's portfolio. I continue to stay involved in AI research.

Work

  1. KeyBank · Lead Quantitative Analytics Associate

  2. Vantura · Founding Engineer

  3. Case Western Reserve University · Research Assistant

  4. Clarity · Growth

  5. KeyBank · Quantitative Analytics Intern

  6. MISCO Refractometer · Engineer Intern

Research

  1. Perpetual Text Generation

    Research on extending LLM generation past the natural stopping point. Introduces a method that detects emergent EOS tokens, backsteps, and applies dynamic temperature adjustment — doubled for exploration, then annealed back to baseline. Pushes Llama2-7B-chat ~1,000 additional coherent tokens past where it would otherwise halt.

  2. Q-Learned CNN Branch Selection

    Reinforcement-learning agent that dynamically selects convolutional branches at inference time, trading FLOPs for accuracy on a per-sample basis. Trained with Q-learning over a branching CNN backbone.

Projects

  1. Vantura

    Get a head start on your mornings with a hyper-customized morning briefing. Receive a carousel on news, sports, and stocks you follow, with an audio podcast that you can listen to on the go. Backed by ElevenLabs, Founder University, and CWRU.

  2. iLiterate

    A web-based language learning platform for second-generation readers rebuilding literacy in their native language. Personalized lessons, flashcards, and a speed reader powered by OpenAI, Gemini, and ElevenLabs on a Next.js + Supabase stack.

  3. XRDimage

    Python package that identifies alpha and beta rings in noisy X-ray diffraction images so researchers can run downstream analysis. Published to PyPI.