Toolkit
This is the actual inventory behind the consulting and the ventures: which models do what, what the agents automate, and what rails this site runs on. Every entry is in live use. Two tools on this page run right now, in your browser.
Live AI surfaces
5
Guardrails
On every route
Stance
Human presses send
The bench
Grouped by layer, one line each on why it earns its slot. If something leaves the stack, it leaves this page.
Models & assistants
The reasoning layer. Each model has one job it is demonstrably good at.
DeepSeek V4 Flash
Runs the public chat and the lab tools on this site. Conversation streams; tools get strict JSON mode. A prefix-cached knowledge pack keeps replies fast and grounded.
Claude (agent mode)
The heavy-lift agent behind the scenes: coding sessions, operations work, and scheduled jobs that run across the ventures before the day starts.
ElevenLabs
When it is switched on, Ask Aziz answers out loud in a cloned voice. The voice layer carries its own rate cap.
Automation & agents
The connective tissue. Tools an agent can pick up without a human copy-pasting between tabs.
MCP servers
Web research (Firecrawl), WhatsApp, YouTube, image generation (ComfyUI), and the X/Twitter API. Each is a tool an agent can call inside one working session.
Scheduled agents
The morning starts with work already summarized instead of a blank inbox, because these jobs brief, draft, and file on a timer.
Site & delivery
The rails this site runs on. Boring on purpose. The drama budget is spent elsewhere.
Next.js 16 + React 19
One codebase serving Arabic-first RTL and English LTR, statically generated wherever possible.
Supabase
Analytics, newsletter, saved workflows, client intake, and the vector store behind the assistant’s retrieval.
Resend
Newsletter sends and lead notifications. Every unsubscribe link is signed and works in one click.
Netlify
Deploys, plus the edge layer that holds the rate limits in front of the AI routes.
Working ops
Where the non-code work lives, so nothing important survives only in someone’s head.
Asana
The task spine across the ventures. Capture first, carry nothing.
Content pipeline
A planning layer that feeds the brands’ channels, with AI drafting inside it and a human pressing publish.
Proof
Five AI surfaces run on azizme.com right now. Same posture as client work: small, guarded, and honest about what the model does. The per-minute caps below are the real production numbers.
Ask Aziz
Bottom corner, every page
Streaming bilingual chat trained on the site’s thinking: persona, knowledge pack, and retrieval over published content. Lives in the corner of every page.
12 req/min
ksa-hosting-compliance
Answers hosting-compliance questions from dated, approved sources only. Every answer carries citation tags and a feedback loop.
12 req/min
projects/deep-research-prompt-maker
A strict-JSON guided dialogue that turns a vague topic into a model-ready research prompt, in Arabic or English.
8 req/min
lifelog-simulator
A live taste of the Work OS on safe demo data: pick a Saudi work scenario, and the model generates the deliverables.
12 req/min
toolkit · prompts
The two newest bench tools. One reads a workflow and names the bottleneck (below on this page); the other reworks rough prompts on the prompt bank.
10 req/min
Every AI route on this site is rate-limited per IP, byte-capped, and validated at the boundary. Same discipline the consulting sells.
Try it live
The same first pass a consulting engagement starts with, compressed into a tool: name the bottleneck, order the moves smallest-first, and say what not to buy yet.
Describe one workflow and where it hurts: intake, follow-ups, reporting, anything. You get the bottleneck named, three vendor-neutral moves ordered smallest-first, and the one thing not to buy yet.
AI output: apply your own judgment before acting on it. Nothing you type here is stored on this site.
Selection rules
Vendor-neutral on purpose. The stack above is what I run; what fits a client depends on these three tests, not on my defaults.
Total cost
Own it end to end
Licence plus the hours it quietly consumes. A cheap tool that needs a babysitter is expensive.
Compliance
PDPL & residency first
Where the data lives and who can see it. Decided before features, not after.
Team fit
Works inside their day
If it needs new habits from everyone, it loses to the tool already open in their tabs.
Packaged assets
The non-software layer: checklists and rhythms that have already survived live engagements.
A repeatable QA and trust pass for AI-assisted products before they go live.
Request accessA scorecard for tracking activation, trust signals, and decision clarity after launch.
Request accessA cadence for design, research, and product syncs that keeps the week moving without noise.
Request accessNext step
A short note on where the team is stuck is enough. I will point you to the right tool, or tell you honestly that you don’t need one yet.