MK

Find a thread

Archive index

24 entries

↑↓ explore Enter open

CV

Software Engineer & Research-Oriented Builder

MohammadKhalife

I turn ideas into working software.

I work across distributed systems, modular-robotics simulation, IoT integrations, and full-stack applications. I enjoy turning complex ideas into practical, testable software.

Distributed SystemsIoTSimulationApplied AI

The collection

Systems, in context.

Selected projects
from code to context.

TypeScript · Next.js · E-commerce

Completed

Ali’s Store — Bilingual E-commerce

An English–Arabic e-commerce platform for a clothing retailer, connecting the online storefront with the store’s day-to-day operations.

Built with H3M Softwares

TypeScript · Next.js · E-commerce

Ali’s Store — Bilingual E-commerce

Ali’s Store brings the customer storefront and the retailer’s back office into one English–Arabic e-commerce platform.

  • Shopping: English and Arabic storefronts, product discovery, size and color variants, favorites, customer accounts, guest checkout, and order tracking.
  • Retail operations: catalog and stock administration, order fulfillment, cash-on-delivery collections, returns, and refunds.
  • Sales and customers: promotions, coupons, volume and bundle pricing, customer records, and loyalty rewards.
  • Administration: business analytics, staff roles and permissions, store settings, email tools, notifications, and database backup and restore.

Technologies

  • TypeScript
  • Next.js
  • React
  • Express
  • Prisma
  • PostgreSQL
  • Redux Toolkit
  • TanStack Query
  • React Hook Form
  • Zod
  • Vitest
  • Docker

Flutter · Android · Local-network monitoring

Completed

Solar Inverter Monitor

An Android app for keeping track of a solar power system, with local inverter monitoring and alerts when grid power goes out or returns.

Flutter · Android · Local-network monitoring

Solar Inverter Monitor

Solar Inverter Monitor brings solar generation, battery state, and grid availability into one Android app. It connects directly to Solarman-compatible loggers through deye_solarman, without a cloud account. Multiple loggers can be managed from the same app, with on-device grid-event history and optional background alerts for power outages and restorations.

  • Connections: network discovery, manual setup, and saved profiles for multiple loggers.
  • Monitoring: live solar output, battery charge and power flow, grid status, and automatic or manual refresh.
  • Alerts: optional Android background monitoring and notifications for grid outages and restorations, controlled separately for each logger.
  • History: on-device storage of grid events, with timestamps, event-type filters, and history management.

Technologies

  • Flutter
  • Dart
  • Android
  • SQLite
  • Riverpod
  • SharedPreferences
  • deye_solarman
  • flutter_background_service
  • flutter_local_notifications

Laravel · Livewire · Student rides

Completed

Lawen Wasel — Student Ride Sharing

A backend and administration platform for student transport, connecting passenger bookings with drivers’ scheduled trips to and from educational institutions.

Laravel · Livewire · Student rides

Lawen Wasel — Student Ride Sharing

Lawen Wasel coordinates student commuting, from finding a suitable trip to managing bookings and the journey itself. Its Laravel API supports passenger and driver workflows, including recurring trips and outbound and return journeys. A Livewire administration application brings transport records, driver verification, and route planning into one place.

  • Passengers: trip search by institution, travel time, and seat requirements; ride requests, driver offers, bookings, and cancellations.
  • Drivers: vehicle management, recurring schedules, request and offer handling, and trip start and completion workflows.
  • Journey planning: separate pickup and drop-off stops for outbound and return trips, Google Maps route views, and OpenRouteService stop ordering.
  • Administration: an operational dashboard, driver verification, passenger and transport records, locations, ratings review, and separate administrator access controls.

Technologies

  • Laravel
  • PHP
  • Livewire
  • Blade
  • Tailwind CSS
  • JavaScript
  • Eloquent
  • Sanctum
  • SQLite
  • Google Maps
  • OpenRouteService
  • Vite
  • Pest

Dart · TCP/UDP · SolarmanV5

Completed

deye_solarman

A Dart library connecting solar-monitoring applications to Deye inverters through local Solarman loggers, without a cloud account.

Dart · TCP/UDP · SolarmanV5

deye_solarman

deye_solarman provides the local communication layer between a Dart or Flutter application and a Deye inverter connected through a Solarman-compatible logger. Adapted from Jonathan McCrohan’s pysolarmanv5, it handles logger discovery and read-only register access through SolarmanV5 and Modbus RTU. It is used in Solar Inverter Monitor.

  • Discovery: UDP scanning across local network interfaces, identifying loggers by IP address, MAC address, and serial number.
  • Communication: direct TCP connections and Modbus holding-register reads wrapped in SolarmanV5 frames.
  • Inverter data: named registers for solar production, battery charging and discharging, and grid status, with signed battery-value decoding.
  • Integration: an asynchronous API for Dart and Flutter applications.

Adapted from jmccrohan/pysolarmanv5.

Technologies

  • Dart
  • UDP
  • TCP
  • SolarmanV5
  • Modbus RTU
  • Git

C++ · Distributed Algorithms · Simulation

Completed

Distributed Leader Election & Shape Recognition

Distributed leader election and relative-coordinate propagation for shape matching in VisibleSim.

Simulation project

C++ · Distributed Algorithms · Simulation

Distributed Leader Election & Shape Recognition

In a separate VisibleSim project, I implemented distributed leader election and relative-coordinate propagation for modular robots to perform shape matching.

  • Implemented distributed leader election across 14,764 simulated modules.
  • Built decentralized relative-coordinate propagation for shape matching.

Evaluated in simulation.

Technologies

  • C++
  • VisibleSim
  • Distributed Algorithms
  • Modular Robotics

Research · Simulation · Distributed Coordination

Distributed exploration and space discovery with 3D catoms.

How can a swarm of modular robots explore a space when it has no map and no target shape? I developed a distributed C++ strategy in VisibleSim that combines growth, branching, and coiling around encountered objects. Each module exchanges information with connected neighbors rather than reading shared state.

Platform
VisibleSim · Self-reconfigurable 3D catoms
Focus
Distributed algorithms · Exploration · Geometry · Simulation
Evaluation
Multi-root growth comparison · Coil-geometry study

Implementation details and source availability are limited by research-sharing constraints.

What the work established

Local rules, coordinated behavior

I combined recruitment, routing, branching, and obstacle wrapping in one distributed strategy, using messages between connected modules.

A comparison of growth strategies

In the tested simulations, growth from multiple roots reduced elapsed simulation time and module motions compared with a single root. I evaluated both time and operational cost.

Geometry that guides the search

I studied how coil radius and pitch affect spaces where an object can remain undetected. The study gives a rule for choosing the search geometry according to object size.

Evaluated in simulation. Physical forces and hardware behavior were not tested; obstacle wrapping was demonstrated qualitatively.

Publication · IEEE MENACOMM 2026

Fine-Tuning and Agentic Workflows for Lebanese Arabizi Understanding and Generation

Co-author of an IEEE MENACOMM 2026 paper on Lebanese Arabizi understanding and generation using open-source large language models.

Read publication

Publication · IEEE MENACOMM 2026

Fine-Tuning and Agentic Workflows for Lebanese Arabizi Understanding and Generation

Abstract

Arabizi—the Latin-script rendering of spoken Arabic using letters, numerals, and punctuation—dominates informal digital communication across the Arab world. Yet most modern Arabic NLP systems assume Arabic-script input, leaving Arabizi largely unsupported. Lebanese Arabizi is particularly challenging due to extreme spelling variability, numerals-as-phonemes (e.g., 3, 7, 8), and systematic trilingual code-switching with English and French.

This work presents the first unified study of Lebanese Arabizi as a conversational interface for large language models. We curate a native-speaker dataset from WhatsApp conversations and develop an evaluation protocol that captures style fidelity, instruction-following, code-switching, and orthographic variability. We compare six modeling approaches: parameter-efficient fine-tuning of open-weight LLMs, a modular Arabizi → Arabic → Arabizi pipeline, managed proprietary fine-tuning, parameter-efficient fine-tuning large models, and an agentic RAG workflow.

Our experiments show that Arabizi-specific fine-tuning substantially improves human-likeness and dialect authenticity over non-adapted baselines. Model scale alone is insufficient for robust Lebanese Arabizi handling without dialect-aware adaptation. The translational pipeline preserves semantic content better but often degrades stylistic naturalness, while agentic RAG improves knowledge-intensive responses when paired with a fine-tuned generator. These results highlight the importance of dialect-aware adaptation and establish a reproducible foundation for Lebanese Arabizi conversational modeling.

Work & research

Experience

  1. H3M Softwares

    – Present

    Freelance Software Developer

    • Part of H3M Softwares, an independent four-developer team working on freelance client projects and software products.
    • Co-developed and delivered Ali's Store, a paid bilingual English–Arabic e-commerce and retail-management platform for a clothing retailer.
    • Currently co-developing a standalone point-of-sale (POS) system as a team-owned product. Development is ongoing.
  2. FEMTO-ST Institute

    –

    Research Intern

    • Designed and implemented distributed C++ algorithms for self-reconfigurable 3D modular robots (catoms) in VisibleSim.
    • Developed coiling strategies inspired by the dodder plant, enabling modular swarms to explore and occupy constrained spaces through local interactions.
    • Evaluated module propagation, geometry adaptation, and space discovery through simulation experiments.
  3. Zaza Woods

    –

    Automation & ERP Developer (Freelance)

    • I developed custom Odoo modules and automation scripts to synchronize inventory and operational data with Shopify.
    • I automated data-entry and reporting workflows using programmed solutions and Excel VBA.
    Work notes

    As a freelance Automation and ERP Developer for Zaza Woods, I developed custom Odoo modules and automation scripts to synchronize inventory and operational data with Shopify. I also automated data-entry and reporting workflows using programmed solutions and Excel VBA.

    • Developed custom Odoo modules and automation scripts.
    • Synchronized inventory and operational data with a Shopify storefront.
    • Automated repetitive data-entry workflows.
    • Built reporting and analysis workflows using programmed solutions and Excel VBA.

    Odoo · Shopify API · Python · Automation · Excel VBA

    Client-specific implementation details and operational data are confidential.

The toolkit

Ideas need instruments.

Programming

  • C++
  • Python
  • PHP
  • Dart
  • TypeScript
  • JavaScript
  • SQL
  • Bash

Systems & Simulation

  • Distributed Algorithms
  • Modular Robotics
  • VisibleSim
  • Linux
  • Docker
  • Git

Web Development

  • Next.js
  • React
  • Node.js
  • Express
  • Laravel
  • Livewire
  • Redux Toolkit
  • REST APIs
  • Prisma
  • Odoo
  • Shopify API

Mobile & IoT

  • Flutter
  • Android
  • Modbus RTU
  • SolarmanV5

Databases & Testing

  • PostgreSQL
  • MySQL
  • SQLite
  • Vitest

AI & Data

  • PyTorch
  • TensorFlow
  • Hugging Face
  • spaCy
  • XGBoost
  • LightGBM
  • Pandas

Education

  1. 2025–2026

    Graduated

    Master 2, Internet of Things

    Université Marie et Louis Pasteur
    Montbéliard, France

  2. 2024–2025

    Completed

    Master 1 in Computer Science

    Lebanese University
    Beirut, Lebanon

  3. 2021–2024

    Completed

    Bachelor in Computer Science

    Lebanese University
    Beirut, Lebanon

Certifications & qualifications

  • Amazon Web Services

    AWS Certified Cloud Practitioner

  • Harvard University

    CS50 Python

  • English language proficiency

    IELTS — Overall band 6.5

I build and study systems that coordinate, simulate, communicate, and adapt.