Malik AlNajjar

MicroMobility booking app on a phone: the event picker, with Open Sports Day at Jeddah Corniche Circuit

Live in Production

MicroMobility Platform

Booking, Check-in & POS Platform · PWA

The production platform behind MicroMobility's group rides and events in Jeddah. Riders book in 10 languages with Google or Apple sign-in and get Apple Wallet passes; staff run check-in (including NFC bike tags), point of sale, inventory, refunds and reports from their own phones, kept in sync in real time. PostgreSQL on Supabase with row-level security, deployed on Cloudflare, with a Playwright end-to-end suite in CI.

SupabasePostgreSQLJavaScriptPWACloudflare PagesPlaywrightGitHub Actions

Challenge

Replace paper queues at a real rental operation: bookings, waivers, check-in and payments on busy ride nights, across several staff phones at once, with customer data locked down.

Approach

One vanilla-JavaScript PWA for riders and staff on Supabase: Postgres with row-level security and staff-only functions, realtime updates instead of polling, and a service worker for install and offline resilience. Cloudflare Pages Functions sign the Apple Wallet passes, and every change passes lint, type and translation checks plus a Playwright suite that stubs the database, so tests never touch production.

Impact

In production every ride night: staff check riders in, assign bikes and take cash or card on their own phones, while riders manage their own bookings in 10 languages, including Arabic and Urdu right-to-left.

View Live ↗

EdgeESP32-S3 · load cell + camera
TransportMQTT events · HTTP clips
HubFastAPI · PostgreSQL 16
MLYOLOv12n · LightGBM · pricing rules
DashboardReact · live WebSocket

Senior Capstone 2025-26Private Repo

AUJ / أوج

Smart Retail Shelf System

Dual-sensor smart shelf for small retailers: a load cell and camera on ESP32-S3 publish events over MQTT to a local FastAPI + PostgreSQL hub (29 REST routes, live WebSocket). A fine-tuned YOLOv12n verifier cross-checks every weight event against its camera clip, LightGBM quantile models (q10/q50/q90) forecast per-SKU demand, and a five-rule engine reprices within a ±15% band with a full audit log. Zero PII, fully offline, 143 passing integration tests, 80 SAR per shelf.

ESP32-S3MQTTPostgreSQLFastAPILightGBMYOLOv12ByteTrackWebSocketReact

AUJ dashboard, Behavior tab: recent shelf picks with product, weight change and camera clip — Behavior tab on the live demo rig

Challenge

Give a small Saudi retailer real-time shelf visibility: live inventory, customer behaviour, demand forecasts, and dynamic pricing, on commodity hardware, fully offline, with zero personally identifying data (PDPL-aligned) and an 80 SAR per-shelf budget.

Approach

Each shelf pairs a 10 Hz load cell running a four-state FSM with a camera: confirmed events publish over MQTT while an 8-frame clip uploads over HTTP. The local hub (FastAPI + PostgreSQL 16) cross-checks weight against a fine-tuned YOLOv12n verifier with a frame-diff fallback, tracks aisle behaviour with ByteTrack using transient IDs only, forecasts demand with LightGBM quantile models trained on the M5 dataset, and reprices daily through five rules in strict precedence inside a ±15% band.

Impact

Pick-to-dashboard latency in seconds, every price change audit-logged, sensor disagreements raised as discrepancy alerts, and 143 integration tests passing against a real PostgreSQL instance. All data stays in the store: no cloud, no PII.

CaptureBrowser extension
Layer 1Great Expectations
Layer 2One-Class SVM
Layer 3Qwen 2.5 · Ollama
ResultDetection 23% → 87.5%

GASTAT Finalist

Sadeed / سديد

Data Quality & Validation Platform

Real-time data quality guard for GASTAT field surveys: a browser extension that catches semantic contradictions (a 19-year-old "general manager", a salary with no job) as the record is typed, before submission. Three escalating layers: Great Expectations structural checks, a One-Class SVM statistical gatekeeper, and a local Arabic-capable LLM (Qwen 2.5 via Ollama) that explains each conflict. Detection lifted from 23% to 87.5% (F1 0.829 on real Labour Force Survey data) at over 97% precision, with zero data leaving the government network.

FastAPIRedisGreat ExpectationsOne-Class SVMQwen 2.5OllamaDockerBrowser Ext.

Challenge

Field surveys collect semantic contradictions rigid rule systems cannot see: a 19-year-old "general manager", a respondent with no job but a 12,000 SAR monthly income, work experience exceeding the respondent's age. Unreliable records reach the national data warehouse and are only discovered when re-collection is already expensive.

Approach

A browser extension validates each record live through three escalating layers: Great Expectations structural checks, a One-Class SVM gatekeeper that forwards only suspicious records at near-zero cost, and a swappable local LLM (Qwen 2.5 / Mistral 7B via Ollama, few-shot prompted, full Arabic support) that judges semantics and explains each conflict. Redis queues scale it to thousands of forms a day, and an open REST API integrates with any survey platform without modifying agency systems.

Impact

Anomaly detection rose from 23% with rules alone (F1 0.374) to 87.5% (F1 0.829 on real Labour Force Survey records) at over 97% precision with fewer than 5 false positives. Citizen data never leaves the government network. Qualified to the national GASTAT Hackathon finals.

View on GitHub ↗

SenseLiDAR · RealSense · encoders
Edge AIJetson Nano · ROS
NavigateAutonomous navigation
AssistTravel assistant app
SecureTLS 1.3 · AES-256 · MFA

Top 15 / 500+ TeamsPrivate Repo

Tayseer / تيسير

Autonomous Smart Wheelchair + Travel Assistant App · AVITHON 2025 (GACA)

Autonomous smart wheelchair paired with a travel assistant app, giving passengers with disabilities full airport independence: ROS mapping with LiDAR, a RealSense depth camera, and wheel encoders fused on Jetson Nano edge AI, with smart routing, emergency monitoring, and adaptive communication for deaf and blind travellers, bookable directly from airline websites. Security by design: TLS 1.3 in transit, AES-256 at rest, MFA with role-based access, and auto-deletion of personal data after boarding (ECC-2 / ISO 27001 aligned). Top 15 of 500+ teams at GACA's AVITHON 2025.

Jetson NanoROSLiDARRealSenseEdge AITLS 1.3AES-256Python

Challenge

Airports overwhelm travellers with disabilities: manual assistance is slow, unreliable, or unavailable at peak times, and existing wheelchairs offer no independence, leaving elderly, blind, deaf, and mobility-impaired passengers lost, dependent on others, and anxious about missing flights or medical emergencies.

Approach

As data scientist on a five-member team: an autonomous wheelchair fusing ROS mapping, LiDAR, a RealSense depth camera, and wheel encoders on Jetson Nano edge AI, paired with a tablet travel assistant offering smart routing, emergency monitoring, and adaptive communication, bookable through airline websites. Security by design: TLS 1.3 in transit, AES-256 at rest, MFA with role-based access, auto-deletion of personal data post-boarding, manual override, and offline operation.

Impact

Field research at King Abdulaziz International Airport (interviews with security officers, a counter shift supervisor, and disability support staff) validated the need: assistance buckles at peak hours and many passengers are too shy to ask for it. Ranked Top 15 of 500+ teams nationwide at GACA's AVITHON 2025.

AppFlutter · Mapbox
BackendNode.js · PostgreSQL · Stripe
SensorsAWS IoT Core
MLSageMaker · scikit-learn
GuidesGuide video + GPS → route

Aseer Tabtaker 2025Private Repo

Masra / مَسرَى

Smart Adventure Tourism Platform

Adventure-tourism platform making Aseer's mountain trails safe and bookable: a Flutter + Mapbox app with trail booking, an AI + IoT early-warning system (AWS IoT Core, SageMaker) watching trail conditions, a network of accredited local guides, and a gamified "Aseer Peaks Passport" stamped per completed trail. Its "Peaks Guide" feature lets certified guides film undocumented trails, and AI drafts the optimal route with rest, danger, and service points from video + GPS for expert review. Presented to the Aseer Development Authority at Aseer Tabtaker 2025.

FlutterMapboxNode.jsPostgreSQLFastAPIAWS SageMakerAWS IoT CoreStripe

Challenge

Aseer's hiking sector is held back by trails nobody knows, serious safety gaps (brain injuries account for 74% of hiking deaths, hitting inexperienced visitors hardest), slow mountain rescue on remote routes, and no unified platform connecting accredited guides with tourists.

Approach

As AI developer on a five-member team: a Flutter + Mapbox app backed by Node.js/Express and PostgreSQL, with Python/FastAPI and scikit-learn models on AWS SageMaker. AWS IoT Core sensors feed the trail early-warning system; the "Peaks Guide" flow turns a certified guide's trail video + GPS into a draft route with rest, danger, and service points for expert review; QR entry manages trail access, with Stripe payments and OpenWeatherMap forecasts.

Impact

A three-phase rollout (Abha pilot, Aseer-wide expansion, national integration with Visit Saudi) projecting 100+ local guide jobs within three years, presented to the Aseer Development Authority at Aseer Tabtaker 2025, in a global adventure-tourism market forecast to reach $2 trillion by 2032.

Who I Am

Experience & Leadership

What I've Built

MicroMobility Platform

AUJ / أوج

Sadeed / سديد

Tayseer / تيسير

Masra / مَسرَى

Technical Arsenal

Industry Expertise

Achievements & Impact

Let's Connect

Ready to build something impactful?

Send a Message