Hassan Hajarat

From the tracker on the truckto the model on the GPU.

Backend lead for a real-time IoT fleet platform. CTO of Mektona, building AI for the built environment. Former founder. I design the system, then direct a fleet of AI agents that build, test and ship it with me.

  • Backend Lead, TrackHawk GPS
  • CTO, Mektona
  • Founder, Ba9i (2021–2024)
7,382

commits shipped in 2026

across 16 repositories

84%

of TrackHawk's backend history

joined mid-2024, after the codebase existed

1,338

commits in 90 days on a new AI product

every one of them mine, with my agents

16

products and prototypes this year

IoT, AI, mobile, games, tooling

Where I build now

Backend Lead

TrackHawk GPS

Jun 2024 – Present · Remote (US company)

Lead the backend of a real-time fleet tracking platform built in-house to replace a licensed vendor product.

  • Own the backend end to end: real-time device telemetry, live tracking, alerts and remote device control for B2B fleets.
  • Led the move of existing customers off the licensed platform and onto ours.
  • Brought AI agents into the team's daily engineering, from bug reproduction to review.
  • Author of ~84% of the backend's commit history since joining in mid-2024.
  • TypeScript
  • Node.js
  • Event streaming
  • MQTT
  • AWS
  • Flutter
  • React

CTO

Mektona

Jun 2026 – Present · Amman

CTO of an early-stage company building AI products for architecture, engineering and construction.

  • Took the first product from prototype to production in three months.
  • Own the technology: architecture, security, infrastructure and QA.
  • Run engineering as one person directing agent fleets, with every change backed by tests and evidence.
  • TypeScript
  • Next.js
  • Generative AI
  • Postgres
  • Cloud infra

Previously

Founder & General Manager, Ba9i (Smart Bus LLC) (Dec 2021 – Aug 2024). Founded a school-bus tracking SaaS for private schools in Amman: GPS trackers on the buses, live maps for parents, reports for schools.

I make the calls. Agents run the lanes.

I treat AI the way a pilot treats an aircraft: as the thing that lets one person cover distances that used to take a crew.

I write the specs, set the architecture, and make the calls. Agents implement, test, review and QA in parallel, in isolated worktrees, against real environments.

Models and vendors will change. The leverage won't. That's why the next layer I'm building is local: models on my own hardware, under my control.

Spec

I write it

Architecture call

I decide

Parallel lanes, each in its own worktree and environment

Implementqueued 0%

waiting for the spec

Testqueued 0%

waiting for the spec

QAqueued 0%

waiting for the spec

Reviewqueued 0%

waiting for the spec

Evidence

diffs, logs, screenshots

Ship

I review and merge

16 products this year. 6 of them are on this wall.

  • Agents · DevOps

    Agent bug-fix loop

    Agents that reproduce a reported bug in a sandbox and come back with a fix and the evidence.

  • AI · Education

    Siba

    Photograph a textbook page; get a checked summary and an exam built only from that page.

  • C# · Godot

    Technical Football

    A physics-driven 1v1 football sim. Simulation core in C#, 196 commits in four days.

    View on GitHub

  • Routing · Mobile

    GigRoute

    Route planning for solo couriers: snap the stops, defer any one, watch the route reflow.

  • iOS

    Vow

    An app blocker with locks that genuinely can't be broken once set.

  • Agents · Automation

    Tickets watcher

    A long-running watcher that tracks ticket availability and acts on it.

Local by default

The assistant on this page runs entirely in your browser, on your GPU. Nothing you type leaves your device. It's a small proof of where I'm going: capable models, on hardware I control.

assistant, localno model loaded

$ load assistant

model
Qwen2.5 1.5B Instruct (4-bit)
download
about 1 GB, once; your browser caches it
runs on
your GPU, through WebGPU
sends
nothing. Questions stay on this device

Nothing downloads until you press the button.

Background and toolkit

Education

Bachelor of Engineering (Honours), Software

The University of Sydney, 2016 – 2019

Thesis in object detection and recognition. Distinctions in Data Structures, OO Frameworks, Internet Software Platforms and Software Development Project. Graduate course in Machine Learning and Data Mining.

Languages

  • Arabic (native)
  • English (IELTS 8.0)
AI-native engineering
Claude Codemulti-agent orchestrationspec-driven developmentagent QA fleetsTDD
Real-time & IoT
KafkaMQTTWebSocketsRedisAWS IoT CoreGPS trackers
Product engineering
TypeScriptNode.jsNext.jsReactReact NativeFlutterPostgresMySQLGraphQL
Infra
AWSTerraformVercelCloudflareGrafana / Prometheus