An AI infrastructure design studio. We architect and forge intelligent systems of any scale — from first sketch to production steel.
01The conviction
“The future belongs to those who shape it — we design intelligence that redefines how businesses think.” — Anas M.K Hasiba, Founder of Archforge
Architecture. The load-bearing structure beneath every intelligent system — designed before a single line is built, drawn to hold weight for years.
The act of shaping raw material under pressure into something strong and permanent. Prototypes are heated, hammered, and tempered into production.
From zero to fully realized production infrastructure. Any scale, any starting point. If it can be drawn, it can be forged.
02What we build
The part that actually thinks. We map how your business makes decisions — cause and effect, what to remember, how to judge — and design that structure before building it, so the system reasons about your world instead of guessing. Drawn on paper first, then built exactly as drawn.
Software that does the work, not just chats about it. Small, focused AI workers that plan a task, weigh the options, and carry it out — with a human approving anything that matters, and a clear record of every step.
Answers from your own facts, not the internet's guesses. We connect the system to your documents and data, then check every answer against the source — so it stays accurate and you can trace exactly where it came from.
Built to survive real, daily use. Monitoring, safety limits, cost control, and testing under pressure — the difference between a demo that impresses once and a system your team can trust every day.
P·01Selected work
A growing ledger of production systems — each shown at the stage it is actually at. Running software, real numbers. No mockups.
Bilingual, Arabic-first exam scheduling for Al-Ahliyya Amman University. A messy registrar file goes in; a provably conflict-free schedule comes out — the model only parses, a deterministic solver builds the grid, and a separate verifier proves zero student conflicts before anything is shown.
Built on a challenge from a solution-architecture expert at Cisco — his problem-solving stages, turned into an agent system. An orchestrator routes each network problem across 16 specialists with hard gates in the path: a Critic red-teams every design, a linter validates every config, and a citation-guard blocks invented RFCs before anything ships.
A bilingual, Arabic-first reference for the Claude ecosystem: trust-tiered skills, 167 copy-ready commands, a roadmap builder — kept current by a scheduled research agent that scans GitHub trending and the official repos every single day.
The studio's largest build: a white-labelable, Arabic-first ERP intelligence platform for a real Jordanian group — hotels, dairy, agriculture, education. A 659-node causal-graph brain debates decisions in council, simulates them before you commit, and nests a full ERP inside every company it runs.
Grounded legal research for Palestinian attorneys over 562 statutes of West Bank and Jordanian law. Every claim carries citations and a confidence score, a deterministic grader decides what may be cited — and when the material is thin, the tool refuses and says exactly what is missing.
Arabic-first real-estate discovery for the Amman market. A buyer describes what they want in one ordinary sentence — «شقة للإيجار بحدود ٦٠٠، غرفتين» — and the assistant turns it into a structured query, returns matching properties with 360° walkthroughs, and narrows by asking rather than by making them fill a form.
A study assistant that explains the way a specific teacher explains, from the material that teacher actually uploaded. Voice is measured arithmetically from their own writing, never guessed; cloning anyone else is refused without recorded consent; and the assistant says plainly it is an AI, won't hand over submittable homework, and won't speak for the teacher about marks, deadlines, or the exam.
The studio's own instrument — the reason every system above was drawn faster and truer than it had any right to be. It composes, it remembers, it grades its own work, and it runs entirely on the edge. We show one screen and seal the rest; the parts and the wiring stay in the vault.
That closes the ledger — eight builds, one kept behind glass. The instrument that shapes them does not go on the page.
03How we work
Your systems, data, and constraints — mapped before anything is drawn. We measure the site before we design the structure.
Architecture drawn as specifications: contracts, data models, failure modes. Agreed on paper before it exists in code.
Short heats — working software every week, shaped against the blueprint. You see the metal move.
Evaluations, guardrails, load and security testing. A blade is only a blade after tempering — strength is demonstrated, never assumed.
Documentation, training, and ownership transferred. Built to outlast the builders.
04Try it now · live
Describe your company and the problem you're stuck on. The studio's own system reads it and returns a first-pass architectural brief — which of our four plates apply to you, what each of the five heats would look like on your build, and where it gets hard. No form, no waiting, no salesperson. This is the same reasoning we bring to a survey, run in the open.
Indicative first pass, not a quotation and not a contract — the survey heat is where a brief becomes exact. Your text is sent to an AI model running on Cloudflare's network; if that model is unavailable, we say so on the answer itself rather than passing a canned brief off as a reading of your text. No human has reviewed the output before you see it. The studio does not store what you type, and nothing is saved unless you choose to send it to us. Privacy.
05Who we are
The craft. Archforge sits between two disciplines that rarely meet: the architect's — designing structures that hold — and the smith's — working the material until it is strong enough to trust.
The gap. Most AI projects die between the demo and production. Our entire practice lives in that gap: taking systems from sketch to steel with the same hands, so nothing is lost in translation.
The method. Deliberately small, solo-founded by design: one human architect draws the blueprint and reviews every line forged against it — directing a harness of 130+ narrow, task-specialized subagents across research, drafting, code generation, review, and test-hardening, running in parallel instead of one generalist model working serially.
The handover. No handoffs, no telephone game — one intent, approved by a human at every gate. Every engagement ships with the documentation, tests, and handover your own team needs to own the system. Typical engagements run two to twelve weeks.
The owner. One name is on the registration, the code, and the invoices — and the whole written record behind that name is published, not summarised. Who owns this empire →
06Anas Eye · the vision
Here is what I see coming — and I don't think it's a guess. The distance between an idea and a working system is collapsing toward zero. The teams that own the next decade won't be the ones that use AI. They'll be the ones that build AI that builds more AI — and then have the discipline to get out of its way.
The edge is no longer a cleverer model — everyone rents the same models. It's a system that improves itself: better output becomes better data becomes better output. A flywheel that never stops turning.
One clever assistant is the past. The future is orchestrated swarms — dozens of narrow, expert agents that plan, build, review, and ship in parallel under a single conductor, around the clock, with humans holding the gates that matter.
When models are commodities, the edge is speed — how fast a system can learn, adapt, and ship again. Not a patent you file once. A living stack that never stops improving.
Software stops shipping once. It becomes a thing that is always mid-upgrade, tuned continuously to the person using it. The last version number is the one you never reach.
This isn't theory to me. Inside my own studio, a harness of 130+ narrow subagents — each scoped to one job: research, drafting, code generation, review, or test-hardening — runs in parallel under my direct review. I don't manage a team of people doing tasks. I review a system that builds, and I sign off on what it ships.
“We're not here to keep pace with the future. We're here to forge the metal it's made of.”— Anas M.K Hasiba, Founder of Archforge
07Before you ask
It depends on the structure, and we would rather scope it honestly than post a number that means nothing. Cost tracks the shape of the system — how much of your data has to be surveyed, how many decisions the system must reason about, and how hard the production bar is. Tell us what you're building and you'll get a straight figure with the draft, before any code is written.
Engagements typically run from a few weeks for a single narrow workflow to a few months for a full intelligence layer. We forge in short heats — you see working software every week, not a silent gap followed by a reveal.
Use them — they're good at what they do. They are not infrastructure. A seat licence doesn't reason over your causal structure, doesn't answer from your own ground truth, isn't gated by a test wall, and nobody is accountable when it's confidently wrong. We build the layer underneath the tools: the one your business actually runs on, and the one you end up owning.
Deliberately small — solo-founded by design. One human architect draws the blueprint and reviews every line forged against it, directing a harness of narrow, task-specialised subagents across research, drafting, code generation, review, and test-hardening. No handoffs, no telephone game. When an engagement needs specialist hands beyond that, we say so before we start rather than after.
Yes. Archforge is a registered company, registration number 565045184, on the official government business registry — and operates from Amman, Jordan. The registration plate in the studio section above carries a QR code that pulls the official record straight from the registry, so you can scan it rather than take our word for it.
You do. Every engagement ships with the source, the documentation, the tests and the handover your own team needs to run it without us. A system that dies when the builder leaves was never a system — it was a dependency.
It stays yours. We work under whatever confidentiality terms your side requires, access only what the survey genuinely needs, and hand back or destroy working copies at the end of the engagement. Say the word and we'll sign your NDA before the first call.
Yes. The studio is based in Amman and works remotely across time zones — the process is built on written specifications and weekly working software precisely so that it survives distance.
08Commission
Tell us what you're trying to build — a first prototype, a system that must scale, or infrastructure that has to survive production. We reply within one working day.