Services
Harnesses, review panels, and evaluation methods, designed and built.
Al Marfa works with engineering teams that want AI agents doing real work without lowering the bar. We design the harness, build it in your repository, and leave behind the tests and the review model that keep it honest. Advisory on architecture and modernization continues alongside.
How We Work
Built, not only described
An engagement ends with working code in your repository: skills, hooks, review panels, evaluation scripts, and the tests that prove them. Written recommendations come with it, not instead of it.
Proven on our own products first
Every method we offer has run on our own work first: a complete Qur'an, a 27,000-hadith collection, and the rebuild of this site, before it is offered to anyone else.
Three shapes an engagement takes
Harness design and build. We map what the agent may touch, write the skills and hooks that hold it there, and add the review panel and evaluation gates. Delivered as code with tests, in your repository, under your review.
Review and evaluation setup. For teams already using agents: a second, independent checker; a record of every verdict tied to the exact text it approved; a sampling plan for the human read; and release gates that refuse anything unverified.
Architecture and modernization advisory. Frontend platform structure, boundaries, and governance, now including migrations carried out by agents with test evidence gating each release.
Where We Help
The areas we work in.
Each service area links to its Expertise page, where the thinking and the case studies live.
Agent Harnesses
Designing and building the skills, hooks, tool boundaries, and evidence formats that let an agent work on your repository or content without a person re-checking everything.
Skills and hooks committed to your repository, with tests
Per-task tool and file allow-lists
A required evidence format for every agent run
Multi-Agent Review
Setting up independent checker agents, blind duplicate runs, and tie-breaker stages so a person reviews only genuine disagreements.
A review panel wired into your agent workflow
Comparison and scoring scripts with their tests
A reviewer queue that shows only unsettled items
Evaluating AI-Generated Content
Building the automatic checks, second opinions, verdict records, and sampling plans that let you publish AI-drafted content with a stated error rate.
A record of every verdict, tied to the exact text it approved
A second opinion from a different vendor
A sampling plan, and the confidence it buys
Agent-Driven Modernization
Planning and running framework upgrades and large refactors with agents doing the change and test evidence gating every release.
A phased plan that says which agent owns which files
Red-first tests per phase
A release gate that fails on missing evidence
Agent Operations
Setting up the work plans, isolated checkouts, model choices, hooks, and reporting that let several agents work one codebase in parallel, including overnight.
A work plan that gives each agent its own files
Which model does which kind of task, and what it costs
A morning report written by a script, not a model
Frontend Architecture
Designing frontend platforms, Nx workspaces, component systems, and governance that stay legible to teams and safe for agents to work in.
Boundary and ownership model with enforced lint rules
Workspace and component platform structure
A governance model that fits the delivery path
To talk through where agents could do real work in your system, start a conversation.