Adam Ezzat
Adam Ezzat
Data Scientist. Language Designer. Builder of things from scratch.
The Path Here
I didn’t start in computer science. I started in economics — studying systems, incentives, and how decisions get made under uncertainty. That curiosity about modeling the real world pulled me toward data, and data pulled me toward code.
My first real technical work was in operations analytics — SQL queries, Excel models, the kind of work where you learn that most business problems are really data problems in disguise. From there I moved into data science proper: building forecasting models, automating pipelines, and learning that the gap between “a model that works in a notebook” and “a system that works in production” is where most of the real engineering lives.
At Amazon Lab 126, I ran experiments with audio recording devices and quickly learned that the hardest part of data work isn’t the modeling — it’s figuring out why your data looks wrong. Is it an environment issue? A script bug? A device misconfiguration? Tracing problems through the full pipeline taught me that clean data isn’t a preprocessing step — it’s the foundation everything else depends on.
At Big Data Insights, I went deeper into ML — designing and deploying forecasting systems using ARIMA, SARIMA, LSTM, and eventually N-BEATS neural architectures. I built the company’s core forecasting product and learned firsthand how much of “data science” is really about building reliable, reproducible pipelines that other people can trust.
The Question That Led to CJC-Lang
Somewhere along the way, a question started bothering me: why is reproducibility so hard?
Every data scientist knows the feeling. You train a model, get great results, and then can’t reproduce them three days later. Different library versions, different floating-point behavior across platforms, hash map iteration order changing between runs. The tools we use are powerful, but they weren’t designed with determinism as a priority.
I wanted to understand what it would take to build a computing environment where the same program with the same seed produces exactly the same output — not approximately, not within tolerance, but bit-identical — every single time.
That question became CJC-Lang: a programming language I’m building from scratch in Rust, designed from the ground up for deterministic scientific computing. No external dependencies for core functionality. No hidden sources of non-determinism. Every floating-point reduction uses compensated summation. Every collection is ordered. Every random number comes from a seeded, splittable generator.
What I Bring Together
Data Science & ML
Time-series forecasting, clustering, predictive modeling, neural network architectures (LSTM, N-BEATS, REINFORCE). Professional experience building and deploying models that actually ship.
Systems & Infrastructure
Data pipelines, automation, validation frameworks, database engineering. The unsexy work that makes everything else possible.
Language & Compiler Engineering
Lexer, parser, type system, HIR/MIR lowering, optimization passes, dual execution engines — all built from scratch in Rust. 20 crates, 5,300+ tests, 6.4 MB single binary.
The Full Stack of Understanding
Economics gave me the “why.” Data science gave me the “what.” Systems engineering gave me the “how.” CJC-Lang is where all three meet.
Where I Am Now
Professionally, I’m a Data Scientist at Big Data Insights, building production forecasting systems. Personally, I’m deep in the CJC-Lang compiler — pushing toward a stable module system, improving type inference, and writing about the process along the way.
I believe the best way to understand a system is to build it yourself. That’s true for forecasting models, data pipelines, and — it turns out — entire programming languages.