CVEmail me

Product-minded software engineer and founder, working with AI agents.

Turki Alsohaim

Exploring how ideas become useful systems.

A depiction of how I work: each idea passes Discuss, Plan, Execute and Verify before it becomes a system. The blocks are my projects.

Scroll

About

I studied software engineering, trained in data and AI at EY, and co-founded Nasq. Now I build products by directing AI coding agents through AI-HQ, the system I started for that work. I write the spec, set the limits and read what comes back. Nothing counts as done until someone other than the maker has checked it.

AI-HQ

A governed system for building software with AI coding agents. I started it and lead it.

Founder interface

Every session starts in Discuss, where nothing is edited.

8runtime packs, one per tool surface

Claude CodeCodexChatGPT
  1. DiscussThink and research. No edits.
  2. PlanWrite the mission and its limits.
  3. ExecuteRun approved missions only.

Red Zone. Money, secrets and anything irreversible need my approval, every time.

Company brain

Decisions, rules and progress are read before any work.

  • Constitution
  • Decisions
  • Progress
Choose a scope

Write to Product repository Allowed

Write to another scope Blocked

Quality kernel

34 capabilities in 10 bundles; each task loads only what it needs.

The task touches
  • Design direction
  • Anti-slop review
  • Accessibility checks
  • RTL checks
  • Browser QA

Ready: everything this task needs is loaded.

Mission control

Every mission is a contract: scope, budget, a reviewer and a way back.

42schemas define its contracts

Asked for The approved model for this task

Answered by The same model

Pass: the approved model answered.

Durable truth

GitHub keeps the record. The maker never approves its own work.

488commits in 11 weeks, with 54 reviewed pull requests

  1. Builder opens a pull request
  2. Verifier reviews it separately
  3. Record ties it to the exact commit
  4. Release Manager merges what passed

Projects

Eight products, each shown in its own look.

Venture, 2024

Nasq

Booking and back office for salons and barbershops, with a website for every salon.

Open the case
Venture, 2023

Quizifyit

My first startup idea: upload your study material and get a quiz made from it.

Open the case
Product, 2026

Orbiqal

Verified, bilingual business profiles that AI assistants can read, for Saudi businesses.

Open the case
Product, 2026

LaunchGate

Launch-readiness checks for Saudi and GCC founders: product evidence in, a score, risks and a fix plan out.

Open the case
Studio, 2026

TORIQ

My studio for websites, AI automation and products, in English and Arabic.

Open the case
Client work, 2026

Adventevo

A paid business assessment: 100 questions across eight areas, a fixed scoring engine and a bilingual report.

Open the case
Client work, 2026

Omorfi

An Arabic-first website and booking system for a women’s salon.

Open the case
Product, 2026

VERT

A personal command center for money, goals and projects, with an assistant that computes before it speaks.

Open the case

Lab

ETH Staking

A testnet validator lab, planned with Codex before any hardware: a 15-point acceptance plan, verified software and a firmware check that fails closed.

PowerShell / Python / Hyper-V / Ubuntu / Nethermind / Nimbus

Clinical AI

A research prototype that drafts reports from sample images, using open-source models on one local GPU.

llama.cpp / MedGemma / Flask / Docker / WSL2

OMNR

Arabic social listening: posts pass a relevance check before AI labels them. Built mostly with Lovable.

React / Supabase / Apify / Claude

Finance Office

A concept for an AI-assisted finance office, sketched as a blueprint: research, one ledger and reconciliation, with a person approving every decision.

Architecture design

Skills and tools

AI agents and models

Data and backend

Interface

Ship and pay

Lab tools

Where it was used

How I work

  1. Clarify

    A clear goal

  2. Write it down

    The plan

  3. Build in small steps

    Tested work

  4. Have it checked

    An independent review

  5. Ship with a way back

    A release with a way back

Experience

  1. EYData and AI trainee2026
  2. MasterworksAI intern2024
  3. Riyad BankSoftware testing intern2023
  4. Prince Sultan UniversityB.Sc. Software Engineering2026

Certified

  • CAPMPMI2025
  • CDMPDAMA2026
  • AI-900Microsoft2025
  • Generative AI and LLMsNVIDIA2026

Email me

Tell me what you’re building.

Send me a message

  1. Your message
  2. This site
  3. Resend
  4. My inbox

0 / 5000

Venture, 2024
  1. A site per salon, on its own domain
  2. Every booking lands in one back office
  3. Guests book with a phone code, no account

Nasqنسق

Booking and back office for salons and barbershops, with a website for every salon.

Role
Co-founder and CEO. I commissioned the product and led an outside team through three milestone deliveries.
Year
2024
Type
Venture

What I built

  1. A site per salonEach salon gets its own website and booking page on its own domain, all from one platform.
  2. Templates on live dataFive templates and a visual editor, connected to each salon’s real services and hours.
  3. Guest bookingGuests book with a phone number and a one-time code. No account needed.
  4. A full back officeServices, branches, calendar, customers, invoices and four reports, on monthly to yearly plans in riyals.

How it works

Each salon’s domain opens its own site, and every booking lands in one shared back office.

  1. A salon’s domain
  2. Its own site
  3. Guest books with a phone code
  4. The shared back office

Built with

  • React
  • Vite
  • Tailwind CSS
  • Fastify
  • MongoDB
  • Docker
  • NGINX
  • GrapesJS
Next caseQuizifyit
Venture, 2023
  1. Your own notes go in
  2. A question drawn from them comes out

Quizifyit

My first startup idea: upload your study material and get a quiz made from it.

Role
Founder. I took it from idea to a public landing page.
Year
2023
Type
Venture

What I built

  1. The ideaYour own material in, personal questions out, at a low cost.
  2. The lessonIt taught me to price the AI before building on it.

How it works

You upload your notes, and every question is drawn from them.

  1. Your notes
  2. Questions drawn from them
  3. Your quiz

Built with

  • HTML5
  • CSS
  • JavaScript
  • Bootstrap
Next caseOrbiqal
Product, 2026
  1. Facts verified against named sources
  2. Published as structured data assistants can read
  3. Re-checked for freshness, never faked

Orbiqal

Verified, bilingual business profiles that AI assistants can read, for Saudi businesses.

Role
Product owner. I wrote the spec pack and directed AI coding agents through the build.
Year
2026
Type
Product

What I built

  1. Spec first25 documents, from business rules to the risk register, written before any code.
  2. Built in eleven daysA monorepo with a web app, a background worker and five shared packages.
  3. Explainable scoringFixed rules decide every score. No AI judge.
  4. Honest statesPayment webhooks are signature-checked, and a check without credentials shows as skipped, never faked.

How it works

Business facts are verified, published with structured data, and re-checked over time.

  1. Business facts
  2. Verified against sources
  3. Public pages with structured data
  4. Re-checked for freshness

Built with

  • Next.js
  • React
  • TypeScript
  • Tailwind CSS
  • PostgreSQL
  • Drizzle
  • Turborepo
  • pnpm
  • Vitest
  • Paddle
  • Vercel
Next caseLaunchGate
Product, 2026
  1. Evidence in
  2. Rules own the score; AI only explains
  3. What blocks launch comes out as a fix plan
  4. The latest payment event wins

LaunchGate

Visit LaunchGate(opens in a new tab)

Launch-readiness checks for Saudi and GCC founders: product evidence in, a score, risks and a fix plan out.

Role
Product owner, with a co-founder. AI coding agents built it; I set the scope and owned the payment decisions.
Year
2026
Type
Product

What I built

  1. Rules own the scoreScore, maturity and access come from fixed rules. AI only explains them.
  2. Payments that fail closedA kill switch, signed webhooks and duplicate-event protection on Paddle.
  3. Caught by a refund testA live refund showed an old “granted” event keeping access open. Now the latest event wins, and a test guards it.
  4. Sources or nothingEvery reference needs a source and a date. Stale ones stop counting.

How it works

Evidence passes fixed rules to become a score. Payments grant or revoke access by the latest event.

  1. Product evidence
  2. Rules
  3. Score, risks and a fix plan
  4. Report, saved or PDF

Built with

  • React
  • Vite
  • Supabase
  • PostgreSQL
  • Paddle
  • Resend
  • Vercel
  • Three.js
  • Framer Motion
  • Playwright
Next caseTORIQ
Studio, 2026
  1. One site, English and Arabic
  2. An assistant, rate-limited per visitor

TORIQ

Visit TORIQ(opens in a new tab)

My studio for websites, AI automation and products, in English and Arabic.

Role
Founder. Built with Lovable from my prompts.
Year
2026
Type
Studio

What I built

  1. Live in four daysFrom template to production in about four days.
  2. Two languages, one adminEnglish and Arabic, with one dashboard for requests, projects, bookings and site content.
  3. An on-site assistantAn AI assistant on the site, served through an edge function and rate-limited per visitor.

How it works

A visitor’s question goes through a rate-limited edge function to the AI model and back.

  1. A visitor’s question
  2. Rate-limited edge function
  3. The AI model
  4. The answer

Built with

  • Vite
  • React
  • TypeScript
  • Tailwind CSS
  • shadcn/ui
  • Supabase
  • Resend
  • Vercel
  • Lovable
Next caseAdventevo
Client work, 2026
  1. 100 questions, saved as you go
  2. Fixed rules score eight areas
  3. Published methods never change

Adventevo

Visit Adventevo(opens in a new tab)

A paid business assessment: 100 questions across eight areas, a fixed scoring engine and a bilingual report.

Role
Delivery owner. I directed AI coding agents in scoped work streams and reviewed each one before acceptance.
Year
2026
Type
Client work

What I built

  1. The journeySign up, pay once, answer 100 questions with autosave, then receive a locked report and a PDF in English or Arabic.
  2. Approved test casesScoring follows fixed rules, checked against 15 cases the client approved. A change that breaks one fails the build.
  3. A versioned methodDraft, validate, publish. Published versions never change, so past reports keep their results.
  4. Secured throughoutRow-level security on every table, audited admin actions, and 499 tests in the final release battery.

How it works

Answers are scored by fixed rules across eight areas, and the report is locked once issued.

  1. Sign up and pay once
  2. 100 questions, autosaved
  3. Fixed scoring rules
  4. A locked report and PDF

Built with

  • TanStack
  • React
  • TypeScript
  • Tailwind CSS
  • Radix UI
  • Supabase
  • Paddle
  • Vercel
  • Zod
  • Vitest
  • Next.js
  • Recharts
  • jsPDF
Next caseOmorfi
Client work, 2026
  1. Arabic first, right to left
  2. Several services in one visit
  3. A taken slot is refused before booking

Omorfiأومورفي

Visit Omorfi(opens in a new tab)

An Arabic-first website and booking system for a women’s salon.

Role
Built with Lovable from my prompts, plus my own work on staff permissions, the calendar and booking clash checks.
Year
2026
Type
Client work

What I built

  1. Several services, one visitBook several services in one visit. A database check blocks any clash before the booking is saved.
  2. Staff-aware schedulingEach staff member’s services, hours and time off shape the available slots.
  3. An admin consoleSeparate views for owner, receptionist and staff, with editors for services, gallery and the homepage.

How it works

Pick services, then a time. The clash check runs before anything is booked.

  1. Pick services
  2. Pick a time
  3. Clash check
  4. Booked

Built with

  • Vite
  • React
  • TypeScript
  • Tailwind CSS
  • shadcn/ui
  • React Query
  • Supabase
  • Vitest
  • Lovable
Next caseVERT
Product, 2026
  1. Routed by the kind of question
  2. Numbers computed in code, first
  3. Memory you can read and delete

VERT

A personal command center for money, goals and projects, with an assistant that computes before it speaks.

Role
Product director. Built with Lovable from my prompts.
Year
2026
Type
Product

What I built

  1. Two lanesAnalysis questions go to Claude, everything else to a faster model. Deep Analysis raises the budget.
  2. Numbers firstProfit and loss, currency and goal progress are calculated in code, then handed to the model with one rule: never invent numbers.
  3. Memory you controlUp to three facts saved per exchange, each one visible and deletable.
  4. Honest statusFour plain states: Healthy, Needs data, Needs verification or Unavailable.

How it works

Each question is routed by kind. The numbers are computed first, then the model explains them.

  1. A question
  2. Routed by kind
  3. Claude, for analysisA faster model, for the rest
  4. Numbers computed in code
  5. Explained by the model

Built with

  • Vite
  • React
  • TypeScript
  • Tailwind CSS
  • Supabase
  • Claude
  • Resend
  • Lovable
Next caseNasq