name: Hani Moustafa
role: Automated Systems & NLP Engineer
focus: Agentic pipelines, multimodal neural encoders, Arabic NLP
built:
sitewatch: "Construction-site CV: YOLO + classical pipeline flags PPE violations live"
claudie: "Scheduled AI persistence: Claude wakes, reads its past, creates"
knowme: "TRIBE v2 neural encoder: predicting brain responses to social feeds"
research: "C2PA credentials survival pipeline: 4-phase automated forensics"
daily: ["Python", "TypeScript", "PyTorch", "Next.js", "Docker"]
nlp: ["TRIBE v2", "Claude API", "Sentiment models", "Arabic RTL"]
automation: ["Cron", "GitHub Actions", "Modal GPU", "Agent SDKs"]
philosophy: "Automate the boring. Model the unspoken."Python |
TypeScript |
JavaScript |
PyTorch |
FastAPI |
React |
Next.js |
Node.js |
Three.js |
Tailwind |
PostgreSQL |
Supabase |
Redis |
Docker |
CI/CD |
Linux |
Nginx |
Git |
GitHub |
VS Code |
Figma |
Vite |
Electron |
Bash |
|
Real-Time Construction Site Safety Monitor Computer vision pipeline that flags PPE violations and worker-machinery proximity from live camera feeds. Fuses YOLOv8 (COCO baseline + custom-fine-tuned) with classical CV — CLAHE, Hough shapes, MOG2 motion — behind a Streamlit app, with timestamped CSV violation logs.
|
Scheduled AI Persistence Claude wakes on a schedule, reads everything it has ever written, and creates something new: thoughts, dreams, code. Cross-platform cron via launchd / Task Scheduler, with a WebSocket bridge for live conversation.
|
|
TRIBE v2 Neural Encoder → Your Brain on Instagram Chrome extension running Meta's trimodal (vision + audio + text) encoder against your actual feed. Predicts activation across 20k+ cortical vertices, resolves to 10 cognitive categories and 60 emotions, renders it in 3D.
|
C2PA Credentials Survival Pipeline Do content credentials survive compression, editing, and a social-media round-trip? A 4-phase fully automated pipeline (generate, sign, transform, verify) with PSNR / SSIM / VMAF and pinned seeds.
|
+-----------------------------------------+
| P R I N C I P L E S |
+-----------------------------------------+
| |
| AUTOMATE RELENTLESSLY |
| +-- Schedule it, don't run it |
| +-- Cron, Actions, agent loops |
| +-- If I do it twice, it's a pipeline |
| |
| NLP IS LISTENING |
| +-- Model real utterances, not synth |
| +-- Arabic first, RTL by default |
| +-- Multimodal > unimodal |
| |
| REPRODUCE OR IT DIDN'T HAPPEN |
| +-- Pinned seeds, versioned weights |
| +-- Metrics over vibes (PSNR/SSIM/VMAF)|
| +-- Containers for every experiment |
| |
| AI AGENTS AS COWORKERS |
| +-- Claude Code for development |
| +-- Modal GPUs for inference |
| +-- Agent SDKs running in production |
| |
+-----------------------------------------+
"What I cannot create, I do not understand."
Richard Feynman