Engineering

Amplify · 2015

The bridge out of the corporate world

My first stop was Amplify, a K-12 education company betting on a classroom tablet of its own. The classroom chose the iPad instead, and that whole product world was gone almost overnight. Then a marketing firm asked me to help with their websites - and their clients started asking too. Nobody ever called it a business plan. It became one.

ctype html> Joshua Lazoff - AI engineer · a story with a supercomputer

Miami, FL · AI engineer · open to the right team · this is a true story

Joshua Lazoff

I am

Twenty years of production engineering for names you know - Goldman Sachs, the team behind MetaMask, Chainlink. It all led somewhere unexpected: a handmade supercomputer on a shelf in my house. Four short chapters tell how.

Senior AI software engineer · knowledge graphs · grounded, cited generation · evaluation pipelines · self-hosted inference

Python · TypeScript · Next.js · Neo4j · Postgres · vector search & embeddings · Playwright · Ollama on a home GPU cluster · Cloudflare edge

What I build

Three systems, live today

A knowledge graph you can talk to, an evaluation harness you can rerun, and a fleet with a company attached. All of it runs on hardware I own.

Maps of knowledge

A three-thousand-year library, alive

A living map of an ancient library - every teaching joined to its author, every author joined to the teachers who taught them, across three thousand years. Ask it anything, and the answer comes with a reference you can look up.

Live now as OpenRabbi - a speaking agent grounded in the graph, running from my house.

Under the hood: a Neo4j knowledge graph, retrieval grounded in the graph, and speech served from my own GPUs.

Evaluation · testing

Measured like production

Written tests grade the system before anything ships, and again in live conversation - references checked against the real text. You can read the methodology, run the queries behind every number, and check any figure yourself.

Python evaluation pipelines, reproducible from the paper: the same queries that grade the system regenerate every figure.

Infrastructure · product

A fleet, run by one person

Dozens of live sites built, deployed and QA'd by pipelines I wrote, on a GPU cluster in my house. Out of it grew Digital Twin Pro: a private AI appliance you buy once and own outright.

Playwright audits gate every deploy: contrast, meta, links and copy are measured on the live page, in both colour modes, before and after every ship.

The personal research behind all of this lives at /torah/.

The story

Four chapters, all true

Every engineer's career is a story about what they were trusted with. Here is mine, told the way it happened. The thread underneath it has its own page.

  1. Chapter One · The boy who fixed computers
  2. Origins

    San Juan, Puerto Rico

    A kid, a screwdriver, a broken machine

    In San Juan, before cloud-anything, computers broke - and stayed broken, unless somebody opened the case. By fourteen, I was that somebody, and grown-ups were paying me. My family were business people, Cuban refugees who had rebuilt from nothing. I was the one happiest with the side panel off, already carrying the question that would run the whole story: how do you fit the real world inside a computer?

  3. Origins

    Ithaca, New York

    A letter arrives

    Nobody in my family had considered a place like Cornell - we did not know to aim that high. The letter came anyway. Information science turned out to be an entire discipline built around my question, and I took two degrees from it (BS ’12, MEng ’13). I am still answering that question today, at larger and larger scale.

  4. Chapter Two · The tower with a secret language
  5. Engineering

    Goldman Sachs · summers 2010–2012

    Three summers, one team

    The same trade-processing team took me back three summers running - a standing invitation, which in that tower is a letter in itself. Inside lived a legend: SecDB, the bank's in-house risk engine, famous for letting the firm see around corners - written in its own private language, Slang, which exists nowhere else on earth. I learned to speak it.

    As an intern I built a reconciliation engine for the new post-crisis regulations, repaired the tool that watches every trade, and wrote an algorithm whose entire job was to watch SecDB itself. It detected changes to the foundations of the machine everyone else trusted blindly.

  6. Engineering

    Goldman Sachs · 2013–2015

    Where correctness is the whole job

    Then the floor itself: Java by day, Slang underneath. I built a query service the whole firm shared and kept hundreds of long-running queries alive at once. I helped stand up a round-the-clock operation too - a hundred-plus people across dozens of teams - that catches errors before they become anyone's worst day. When your software sits between a bank and its regulators, correctness is not a virtue. It is the whole job.

  7. Chapter Three · The new world
  8. June 2015

    Two companies, one month

    I left the “perfect job” and incorporated two companies in the same month: LazoffTech, to earn - and TorahTech, for the long project underneath, the one this story has secretly been about since the beginning. The first has funded the second ever since. Within a year I was engineering a digital home for Jewish learning by day (Project Mercava) and writing myself a five-line plan titled “Websites to make.” The first line was this website.

  9. Engineering

    ConsenSys · five years

    The front door to a new financial world

    I joined through the education arm, teaching developers an entirely new way to program, and rose to lead software engineering manager. That put me over every website behind MetaMask and consensys.io - the pages millions of people walked through into an economy that had not existed a decade earlier. Sub-second page loads, measured; a 99% technical SEO score; the web presence of software that guards people's money. That teaches you a professional level of paranoia. I recommend it.

  10. Engineering

    Chainlink · three and a half years

    The company whose product is an oracle

    Chainlink builds the systems that tell blockchains the truth about the outside world - prices, events, facts. The industry's own name for them: oracles. I built the sites and tools its developers rely on - three of those years while still running the web at ConsenSys, both engagements flowing through the research company. The quest was being funded, even when it looked like a day job.

  11. Chapter Four · The machine in the house
  12. Engineering

    Miami · 2023

    The handmade supercomputer

    Then the models arrived, and I understood what twenty years had been for. I stopped renting my computing and built my own: GPUs racked one at a time on a shelf in my own house, storage I own, models I can open and inspect. Not in a data center. In the house - where the important libraries have always been kept.

  13. Research

    Now

    Where the oldest library meets the newest mind

    On that machine, the two threads of this story finally merge. I turned the oldest library I know - the Torah and its commentaries - into a living map: every teaching joined to its author, every author joined to their teachers, three thousand years of conversation made navigable. Eight voices answer from it out loud, and anyone can talk to them. Everything else I run grew up around this machine.

  14. The next chapter

    Every story needs one more impossible thing

    The boy who fixed computers has a fleet now, and it runs in public. What happens in the next chapter depends partly on who is reading this. If you are building something that matters - AI people have to be able to trust - then the next chapter might be ours. Write to me.

Built and running

The sites

49 live sites. Every one of them started as something I wanted to understand well enough to explain out loud. Half my career happened in classrooms nobody called classrooms - teaching calculus, mentoring interns, building learning platforms. These sites are that same instinct with better tools, and every one of them is open to you. Here are six to start with; the rest, with what each one is for, live at jew.tech.

Connect

One conversation I always take

Are you building AI that people have to trust? Do you need an engineer who ships, measures, and publishes the results?

I am looking for my next role: AI engineering, knowledge systems, production intelligence. That is the conversation I always take. Write to me and I will answer - with sources!

Want the full picture first? The research is at jew.tech, the receipts at jew.tech/evidence, and the experiments I want to run next at jew.tech/agenda.