Services
Five ways we put AI to work — and one that tells you not to.
Every engagement starts the same way: what is the job, who does it today, and what would have to be true for a model to do it better. The answer decides which of these you actually need.
Service
AI agents & automation
Tooling · Evals · Guardrails
An agent is only useful when it can touch your real systems. We build ones that read from your database, call your internal APIs, and return work in a form a person can approve or reject — with every step logged.
The hard part is not the model. It is deciding what the agent is allowed to do unsupervised, what it must escalate, and how you find out when it is quietly wrong. We design that boundary before we write the prompt.
- Typical use Document processing, back-office triage, customer support escalation, internal research
- Included Tool integrations, permission model, evaluation harness, human-in-the-loop review UI, audit log
- Runs on Claude, GPT, Gemini, or an open model you host — chosen after testing, not before
Service
LLM product engineering
RAG · Arabic NLP · Fine-tuning
Copilots and assistants that answer from your own material — contracts, policies, tickets, product catalogues — instead of from the open internet. Retrieval done properly, so the answer comes with a citation you can open.
Arabic is where most retrieval systems break: mixed script, dialect, inconsistent transliteration, and PDFs that were scanned rather than typed. We treat Arabic as a first-class requirement, not a translation layer added at the end.
- Typical use Internal knowledge assistants, contract and policy search, support deflection, structured extraction
- Included Ingestion pipeline, chunking and embedding strategy, citation UI, hallucination testing, cost-per-query budget
- Languages Arabic, English, and mixed-script queries in the same index
Service
Data & ML platforms
Pipelines · Serving · Monitoring
Most stalled AI projects are not model problems. They are data problems wearing a model costume — the numbers do not reconcile, nobody owns the pipeline, and no one can say where a figure came from.
We build the layer underneath: ingestion, transformation, storage, serving, and the monitoring that tells you when quality drifts. Boring, and the reason the interesting part works.
This is not a pivot for us — data mining and CRM work have been on our services list for years. Machine learning is what we point that same pipeline at now.
- Typical use Forecasting, scoring and ranking, recommendation, anomaly detection, reporting that reconciles
- Included Pipeline orchestration, model serving, drift and cost monitoring, dashboards, runbooks
- Hosting Your cloud account, in-region where residency matters
Service
AI-enabled web & mobile
Web · Mobile · Integrations
Complete products, built the way we have always built them — marketplaces, payment flows, dashboards, operations tools — with the model work designed in from the first wireframe rather than bolted on after launch.
This is the oldest thing we do, and the work behind every client on our projects page. The discipline that keeps a booking system upright is the same discipline an AI feature needs.
- Typical use Customer platforms, marketplaces, internal operations tools, mobile apps
- Included Product design, front end, back end, integrations, RTL and Arabic interfaces, accessibility
- Handover Your repository, your infrastructure, documentation your team can follow
Service
AI discovery sprint
2 weeks · Fixed price
Two weeks, a fixed fee, and one question: is AI the right tool for this problem? We map the process, build a prototype against your real data, and measure it against how the job is done today.
"No, not yet" is a valid outcome and you still keep everything — the prototype, the evaluation set, and a costed plan for the build if the answer turns out to be yes.
- Week one Process mapping, data review, feasibility, cost modelling
- Week two Working prototype on your data, evaluation set, results readout
- You keep The prototype, the evals, the cost model, and a build plan — whatever we conclude
Next step
Not sure which of these you need?
Describe the problem instead of the solution. We will point you at the right one — including at nothing, if that is the honest answer.