Full-stack and AI engineer
I build real products and ship them: GourmetAI, an AI that turns phone photos into studio shots; eSIMs5G, a travel eSIM app in both app stores; and Ka2s menus, restaurant menus in three languages.
● Open to full-stack and AI roles · Israel · Arabic, Hebrew, English
How I work, then four projects, live or on GitHub.
How I work
One early batch read 739 million tokens, so I measure my usage and cut waste: one big model plans, small ones do chores.
Claude Code
GitHub
Gemini
Python01 / 04 Live
Turns a phone photo of a dish or product into a studio-style shot, for restaurants and shops without a photographer.
React
TypeScript
Node.js
PostgreSQLGemini
02 / 04 Live
Travel data for 190+ countries. Buy an eSIM in the app, install it with a QR code, and land online.
React Native
ExpoTypeScriptNode.js
Supabase
03 / 04 Live
Restaurant menus that guests open with a QR code at the table, in one of three languages, right to left included. Updated without reprinting.
Next.js 15React 19TypeScript
Tailwind
next-intl
04 / 04 Live
A GPT-2 that searches the Game of Thrones books and answers from what it finds. I wrote its network in PyTorch, following a build-an-LLM course, and fine-tuned it on the books.
Python
PyTorch
BM25
FastAPI
ONNX Runtime
Research, AI experiments and smaller builds.
Small neural networks compare writing styles to flag ancient Arabic books whose credited author doesn't fit.
M.Sc., 2021–2023Read more
The same GPT-2 network with its 50,257-word output swapped for two: spam or not spam.
A delivery app and server: customers send a location pin on WhatsApp, and the driver gets every stop on a map, in order.
Private app, 2025784 commits
It reads each pull request and leaves comments with fixes. My whole team used it every day.
Internal tool, 2025–2026
A silent film drawn in code: the matrix math inside every GPT, from scores to the output.
Film20 s, no sound
Ten notebooks from the build-an-LLM course: tokenizer, attention, a full GPT, then fine-tuning.
CodeOpen on GitHub
Test automation and data first, then full-stack and AI, now my own products.
Jun 2026 — Present
I built all three alone, end to end.
eSIM app, website and AI admin manager: 500+ B2C users, 70+ B2B travel agencies.
Studio pictures from phone photos; 20 users.
QR menus in Arabic, Hebrew and English for 5 restaurants.
TypeScriptReactNext.jsReact NativeNode.jsPostgreSQL
RedisGemini
OpenAIOct 2025 — May 2026
Full-stack work on an internal enterprise platform for API usage, pricing and payments, plus the team's AI tooling.
TypeScriptReact
Copilot agents
Prompt engineeringMar 2025 — Oct 2025
Seven months building my own products and learning how language models work.
React Native
FlaskPyTorchFeb 2023 — Mar 2025
Built, from scratch, a platform that loads production-shaped data into test environments, and ran it alone for five months.
Python
Pandas
AWS S3
GitLab CI
Salesforce APIsMar 2022 — Feb 2023
Owned the system-test infrastructure: nightly suites, feedback, reports to Jira and Slack.
Python
Selenium
PytestGitLab CIJan 2021 — Mar 2021
Car detection with YOLOv3 and OpenCV, and a TensorFlow classification model.
Python
TensorFlow
OpenCVFinal project: ancient Arabic author verification.
Final project: credit-card fraud detection with Isolation Forest, in a React Native app I taught myself.
Arabic, my first language. Hebrew and English, professional.
I'm Karam. My projects start from a real problem: a restaurant with no budget for a photographer, a traveller who lands without data, a menu in three languages.
I learn by building the thing myself. Away from work I trained a GPT-2 on A Song of Ice and Fire, because I love the books, and I made a Mac tool that opens my work apps when I clap my hands.
Say hello.The fastest way to reach me is email.
Have an access code? Enter it here.
