An open-source, local-first memory layer that gives AI assistants persistent, human-like memory instead of resetting every conversation. Built as a Model Context Protocol (MCP) server, it works with any MCP-compatible model (Claude, GPT, Gemini) and runs entirely offline with zero telemetry. The system uses a layered memory architecture inspired by cognitive science — a Fast Layer for core identity (always loaded), a Warm Layer for context-specific attributes retrieved by semantic relevance, and an Archive layer (SQLite + multilingual sentence-transformer embeddings) for long-term storage — combined with a rule-based forgetting mechanism that scores memories by frequency, recency, and emotional weight, permanently protecting emotionally significant memories while pruning trivial ones over time. To guarantee reliable behavior for non-MCP-native integrations, it includes a deterministic Gateway API that always injects context and auto-stores relevant information, removing the dependency on a model choosing to call a tool. The project is fully documented with an architecture guide, a roadmap, and nine Architecture Decision Records explaining key engineering trade-offs, and is released under the Apache 2.0 license.
whoami
Deeb Hajjar — self-taught software developer from Mount Lebanon, running a Python/Django core with an AI/ML specialization compiling in the background.
I build backends that hold up: REST APIs, PostgreSQL schemas that make sense at 2 AM, payment flows, inventory math. And I build the thing I care about most — machine memory. My Human Memory Project gives AI a five-layer memory (working, episodic, semantic, procedural, personality) with continual-learning consolidation, so the system adapts instead of forgetting.
// the tunnel behind this page is the pipeline. keep scrolling — you're flying through it.
ps aux | grep deeb
lb-commerce.service
RUNNINGA comprehensive Django e-commerce platform built specifically for the Lebanese market, deployed as a single-tenant, self-hosted solution per client. It supports cash-on-delivery only (with a Payment model already in place for adding gateways later), and dual-currency handling — USD as the source of truth, with LBP snapshotted for display at a configurable exchange rate. The platform features a fully bilingual interface (English/Arabic) with complete RTL/LTR support, and a comprehensive admin panel (Site Configuration) that lets the store owner control everything — branding, homepage sections, shipping settings, and feature toggles like inventory and returns — without any redeploy. It also includes an optional inventory module with full stock-movement audit trails and weighted-average costing, along with coupons and a staff analytics dashboard. Built with Django 5.2, PostgreSQL, and Redis, and deployed via Docker Compose behind nginx, with a test suite of 229 tests. // private unit — demo on request
Learning Academy — A comprehensive online learning platform built with Django and Bootstrap 5, featuring course management, a role-based blog system, Stripe payment integration, and multi-language support (English/Arabic with RTL). Developed as a graduation project for Hsoub Academy's Python Web Development course, with cloud storage on AWS S3 and PostgreSQL in production.
Inventory Management System API — A Django REST Framework API for managing products, categories, and stock transactions, with real-time inventory tracking, low-stock alerts, token-based authentication, and Swagger/OpenAPI documentation.
Inventory Management System (Frontend) — A responsive React application consuming a Django REST Framework API, featuring real-time stock monitoring, low-stock alerts, and interactive analytics dashboards (Recharts) with dark/light theming and TanStack Query for optimized data fetching
htop --skills
tail -f /var/log/learning.log
ssh you@deeb --connect
// open to backend & AI/ML roles — remote or Lebanon. channel is encrypted with good faith.