Skip to content
Yuval Gerber

Computer Science · Columbia University

Hi, I’m Yuval Gerber.

A Computer Science student at Columbia and an aspiring AI Engineer. I build AI systems from first principles — like a language model trained from scratch on my own laptop — and real-time software that runs on its own.

RésuméGitHub ↗LinkedIn ↗Email
Yuval Gerber smiling in a snowy New York park, beneath a lit Central Park lamppost.

About

A little about me

I study Computer Science at Columbia University, and I want to become an AI Engineer — someone who understands a model all the way down and can still ship the product around it.

I learn by building end to end: the data pipeline, the model, the evaluation and the app. Before Columbia I served in the Israel Defense Forces as a laboratory commander for Black Hawk avionics, where “it usually works” was never the standard.

Education

  • Columbia University

    B.A. in Computer Science

    Relevant courses: Data Structures · Advanced Programming · Discrete Math · Linear Algebra & Probability · Statistics

  • Central Washington University

    Biology, B.S. program

Based in

New York City

Projects

Selected work

  • 01 — AI / Deep Learning

    MiniGPT Studio

    An end-to-end pipeline that trains a decoder-only Transformer from random weights on Apple Silicon, evaluates it honestly, and lets you talk to it in a native-feeling macOS app. No pretrained weights. No hosted model APIs.

    parameters
    29.4Mparameters
    token corpus
    445Mtoken corpus
    pretrained weights
    0pretrained weights
    test perplexity
    18.54test perplexity
    • Python 3.12
    • PyTorch (MPS)
    • NumPy
    • HF tokenizers
    • PyArrow
    • PyQt6
    • Objective-C / AppKit
    Read the case study →Source on GitHub ↗

  • 02 — Real-Time Systems / FinTech

    HYPE

    A single-user, real-time copy-trading system for Solana. It streams trader activity over WebSockets, screens every token with parallel risk checks, executes and manages its own exits, and runs behind a secured PyQt6 dashboard and Telegram bot.

    lines of Python
    13,927lines of Python
    pre-trade checks
    8pre-trade checks
    live services
    6live services
    win rate
    88%win rate
    • Python 3.12
    • asyncio
    • threads
    • WebSockets
    • SQLite (WAL)
    • PyQt6 + pyqtgraph
    • Pydantic v2
    Read the case study →Source on GitHub ↗

Experience

Where I’ve worked

Electronic Warfare Engineer, Laboratory Commander

2016 — 2019

Israel Defense Forces — Technical Unit

  • —Led the laboratory that tested, diagnosed and repaired electronic-warfare and avionics systems on Black Hawk helicopters.
  • —Built and enforced quality-assurance and safety-compliance procedures under strict time and zero-failure constraints.
  • —Supervised a team of 50 technicians and engineers across testing, repair and quality assurance.
  • —Received the Elective Modification Award — the top distinction among roughly 300,000 soldiers — for extraordinary contribution in system innovation.

Skills

What I work with

Machine learning

  • PyTorch
  • Decoder-only Transformers
  • Training from random init
  • Byte-level BPE tokenization
  • Evaluation & perplexity
  • Leakage-aware data splits
  • Apple Silicon (MPS)

Systems

  • Python · asyncio
  • Threads & process isolation
  • WebSockets
  • REST API integration
  • SQLite (WAL)
  • Fault-tolerant checkpointing
  • Rate limiting & backoff

Product & tooling

  • PyQt6 desktop apps
  • pytest
  • Pydantic
  • PyArrow / Parquet
  • Git & GitHub
  • Spec-first development

Contact

Let’s talk

Internships, research, collaboration or a good technical conversation — send a note.

y.gerber00@gmail.com

LinkedIn ↗GitHub ↗

0 / 5,000

Your name, email and message are sent to me by email and used only to reply. See the privacy policy.