Strategy & model selection
Which model holds quality for your workload, on hardware you can afford. Benchmarked against your data, not a leaderboard.
- Use-case triage
- Open vs. closed evaluation
- Build / buy / wait call
The model is the cheap part.
Independent advisory for companies making their first serious AI investment. A decade of enterprise platform engineering — the infrastructure other teams build on — pointed at what your deployment actually costs to build, run, and keep running.
- Team
- Engineers, SEO experts, photographers
- Credentials
- Ex-Microsoft & Apple, granted U.S. patents
- Reach
- Worldwide · U.S. focus
Practice areas
Where we come in
Before you commit, before you sign, and after it's live.
Cost & capacity modeling
Unit economics at your real volume: where self-hosting overtakes a managed API, the bill at ten times the traffic, and what breaks first.
- Token & GPU economics
- Break-even analysis
- Vendor contract review
Deployment & readiness
Architecture, failover, latency budgets, and the on-call story. The part that decides whether it survives month three.
- Reference architecture
- Observability & alerting
- Runbooks & handoff
Adjacent practice
Digital presence & market share
The line this practice was built on, and still running: web presence, technical SEO, and e-commerce growth. Same discipline — measure it, find the waste, remove it — applied to page speed, crawlability, and conversion. See selected work.
Your spend
What it costs to run
Input and output are priced differently, caching and batching change the maths, and retries quietly inflate every line. Put your real workload in — then see what would actually cut it.
Estimated year-one spend
$0$0 in month one
$0 per request
$0 per 1,000 requests
0% of the bill is output
Your workload
Where the money goes
What would actually cut this
—
Managed API or Self-Host?
—
Input and output are billed separately — output typically runs three to five times
the input rate, which is why a workload that looks cheap on paper isn't. Retries are
applied to both before anything else, because a failed call still bills. Cached input
is charged at the discount you set; Anthropic reads cached tokens at roughly a tenth
of list, other providers differ. Batch and committed-use discounts are applied last,
to the subtotal. Year one compounds monthly growth rather than multiplying month one
by twelve. Self-host break-even is (GPU + ops × $95/hr) ÷ cost per request.
Excluded: engineering time, evaluation, embeddings, vector storage, egress and
the cost of an outage — which routinely exceed the inference bill in year one. Model
prices are indicative list rates and every field is editable; verify with your
provider before committing budget.
The same workload, every model
| Model | In / Out per 1M | Per month | Year one | vs. yours |
|---|
Selected work
Shipped, and still running
Sites designed, built, photographed, and kept running under Daraghmeh Digital.
- Faraci Lange Attorneys Firm site
- International Food Market E-commerce
- RIT Behind the Bricks Social platform
- Daraghmeh USA Company site
- FlickShare Company site
- Arch Hookah Branding site
Design & build
Structure, front-end, and CMS — built to load fast on a phone and stay maintainable after handoff.
Technical SEO
Crawlability, Core Web Vitals, structured data, and the unglamorous fixes that actually move rankings.
Photography & content
Original product and location photography, so the storefront isn't furnished with stock images.
Hosting & maintenance
Updates, monitoring, backups, and a real person to call the day something breaks.
Process
How We Operate
01
First conversation
We ask the questions that expose the real constraint, then tell you whether you need a consultant at all — sometimes the answer is no.
30 minutes · no obligation
02
Assessment
A written deliverable: benchmarked model recommendation, a cost model at your volumes, reference architecture, and the risks nobody flagged. Yours to keep and to shop around.
1–2 weeks · quoted in advance
03
Build & handoff
Implementation oversight next to your team or your vendor, through launch and into steady state — then a handoff package so nothing depends on us.
Monthly · fixed scopeWorking with us
Common Questions
Confidentiality, data handling, and what happens if it goes wrong.
Who sees our data, and what happens to it afterwards?
Only the people assigned to your engagement. Wherever it's technically possible we work inside your environment rather than copying data out, so your systems stay the system of record. Nothing you share is used to train a model, reused for another client, or kept once the engagement closes — and we'll confirm deletion in writing on request.
Can you work under our security and compliance requirements?
Yes. We've delivered on platforms carrying GDPR obligations and regional data-residency rules, and we're used to working inside controls set by someone else's security team — background checks, access reviews, dedicated accounts, restricted environments. Tell us the constraints and we'll work within them rather than asking for exceptions.
What if the work isn't what we expected?
Every engagement runs against a scope agreed in writing before anything starts. If what we deliver doesn't match that scope, we revise it at no additional charge until it does. And if you decide after the first week that this isn't working, you pay for the week that's been worked, keep everything produced to that point, and we part on good terms.
Do you resell anything?
No. No hardware, no licenses, no cloud credits, no referral fees from model providers. The only thing we sell is our time — which is exactly why we can tell you to spend less.
What if we don't actually need AI?
Then that's the answer, and you get it in the first conversation before you've paid anything. A surprising share of "AI problems" turn out to be reporting problems, data-quality problems, or a search index nobody ever tuned.
Why does an AI practice also do SEO and e-commerce?
Because it's the line this practice was founded on and those clients are still served. Most volume today is AI and infrastructure advisory, but the web work draws on the same engineering habits.
How do you charge?
Assessments are a fixed fee quoted after the first conversation, so you know the price before committing. Ongoing work is a monthly retainer against a fixed scope. No hourly billing, and no change order for asking a question.
Contact
Get in touch
The first conversation is short and free. If it isn't a fit, we'll say so and point you at what is.
- Emailmohammad@daraghmehdigital.com
- CompanyLinkedIn
- OwnerMohammad Daraghmeh — www.daraghmeh.me