Independent developer · Lebanon & remote GCC delivery
AI agents with a defined job, useful integrations and clear limits.
I build AI-enabled workflows and agents for teams that need help interpreting information, retrieving answers or preparing actions across their tools. Delivery is remote from Lebanon for businesses in the GCC and elsewhere, with the agent's responsibilities and access defined before implementation.
Discuss your projectShare a short outline. I’ll review the fit, discuss your tools and requirements, then propose a paid scope before implementation.
See my personal RAG and AI-agent toolkit walkthrough →Start with the business problem.
An agent should have a bounded task and a reliable way to check its work. The design needs to distinguish what the model may suggest, what it may execute and what must go to a person. Predictable steps can remain ordinary software rather than becoming model decisions.
What we can scope
- Support triage with retrieval from approved information
- Lead qualification and routing into an existing CRM
- Document classification and extraction with human review
- Tool-assisted internal workflows with explicit permissions
What the delivery can include
- A task definition, input sources and allowed tool actions
- An evaluation set and agreed acceptance criteria
- Escalation, uncertainty handling and approval requirements
- Deployment guidance, usage-cost review and a scoped maintenance plan
Illustrative workflow · not a client case study
Example: support triage with a person in control
An agent can interpret an incoming request, retrieve approved information and prepare a response or tool action. This is an illustrative design to discuss, not a deployed client case study.
- Receive a support request
- Classify the request within agreed categories
- Retrieve approved context
- Prepare a reply or permitted tool action
- Send uncertain or sensitive cases to a person
What needs engineering: Agree which actions need approval, what happens when a tool fails and which examples demonstrate an acceptable result. A model suggestion is not permission to change a business record.
If every decision can be expressed as a stable rule, ordinary automation may be simpler to operate and test. Compare with rule-based workflow automation →
What shapes the project quote?
- The task, allowed tools and systems the agent needs to access
- Representative evaluation examples, failure cases and approval requirements
- Model usage, hosting, logging and maintenance requirements
Scope and boundaries
The agent receives only the permissions agreed for its task. Unsupervised access to sensitive systems, guaranteed correctness and unlimited model usage are not implied. Actions such as sending messages or changing business records need an agreed approval policy.
Is this a useful first project?
An initial project works best when a team can provide representative examples, permission to use the relevant information and a process owner to evaluate results. Unrestricted access to sensitive systems or unsupported accuracy promises should not be part of the starting scope.
Relevant engineering work
Jamil's portfolio includes personal agent tooling and a multi-agent frontend code generator. These show technical work with agent orchestration. Retrieval and tool action are evaluated separately: the RAG knowledge assistants work (see its proof and the personal RAG/AI-agent toolkit walkthrough) demonstrates source-grounded retrieval, not permissioned tool execution; agent development proof above covers orchestration and tool-call structure, not a record of production actions taken on a business's systems. A business deployment still needs evaluation against its own task, data and operating constraints.
Read the Personal RAG and AI-agent toolkit project walkthrough →Explore the project portfolio →Before we start
- How does an agent differ from workflow automation?
- A workflow follows defined steps. An agent may use a model to interpret an input or choose an allowed tool action. Many business systems combine both, with deterministic rules around the model's decisions.
- Can the agent use our business information?
- Approved information can be connected through retrieval or suitable APIs. Access rights, hosting, data handling and what may be sent to model providers must be agreed for the specific project.
- Can you guarantee every answer is correct?
- No. The project should use representative evaluation, clear limits and human escalation for cases that require judgment or cannot be checked reliably.
- Do we need multiple agents or one bounded assistant?
- Start with the smallest design that can be evaluated. Several agents add model calls, handoffs and failure paths; they are useful only when separate responsibilities justify that complexity. A single assistant with controlled tools or a deterministic workflow may fit better.
- What affects AI agent development and running costs?
- The quote depends on integrations, permissions, evaluation, interface and deployment requirements. Running costs depend on model calls, context size, retries, hosting and monitoring. Token usage alone does not estimate development, human review, security work or return on investment.
- Can you work remotely with teams in the UAE, Saudi Arabia and Qatar?
- Yes, from Lebanon. Before scoping, discuss your procurement requirements, working language, account ownership, hosting region and data handling. Country-specific requirements and Arabic-language evaluation need explicit review; remote delivery is not a claim of a local office or automatic regulatory compliance.
Bring a workflow or application brief.
Share your company, current tools and the work you want to improve. Smaller projects are welcome. We can discuss fit, scope, integrations and a quote before work starts.
Describe your project