Independent practice · Saudi Arabia
Defined-scope automation, agent-trace review and workflow integration. Built around a clear problem, inspectable work and agreed acceptance tests.
01 / Services
For operations teams, AI builders and implementation partners who need a concrete technical contribution.
Examine one recurring workflow across up to two systems. Establish the baseline, map data and access requirements, and decide what is worth building.
Starts with a process owner and approved examples. Production implementation is a separate scope.
Discuss an assessmentReview an agreed export of agent execution records for structural consistency, evidence references, observable failures and reproducibility limits.
One normalized schema; up to 100 traces and 10,000 total events within agreed tool limits. Technical review, without a certification or safety score.
Discuss a trace reviewDevelop one agreed Python or API workflow in an authorized sandbox, with explicit interfaces, permission boundaries and failure handling.
Up to two agreed interfaces and one environment. Discovery precedes a fixed-price commitment; production changes are separately agreed.
Discuss an integrationEffort ranges are planning estimates. Schedule and fee follow a review of requirements, sample data and access. External service costs and legal terms are agreed separately.
Public service details02 / Selected work
Research software, a runnable workflow sample and a developing trace-audit tool. Each has a different scope and maturity.
Public research software
01From a frozen evaluation corpus to traceable claims.
Source manifests, record selection, raw-to-derived lineage and provenance-fault checks are linked to reproducible outputs. The v1.0.0 release preserves its declared scope and negative results.
Benchmark-specific research software. External replication and cross-benchmark generalization have not been established.
Synthetic workflow demonstration
02A local Python/SQLite sample that binds approved changes to a specific plan, checks stale input, records transactional changes and handles replay.
Synthetic records and simulated approvals. The public verification record describes the packaged test run; this is a portfolio demonstration.
Prototype
03An offline checker for supported normalized JSON and JSONL traces. It reports structural issues, broken evidence references, declared unknowns and optional local artifact-hash comparisons.
Tested with synthetic fixtures. It does not establish source authenticity, semantic truth or complete CTVI construct validity. Directory publication is not claimed.
03 / Research & collaboration
Research informs the work. Proposed constructs, computational results and commercial tool capabilities remain distinct.
Research into agent-trajectory validity and evidence admissibility: what execution records can support, which evidence is missing, and where a claim must remain uncertain.
Full construct validation and independent external validation remain open. The public trace-audit software is one narrower implementation.
Theoretical work on how machine-mediated coordination may couple operational capability with systemic fragility.
Literature, evidence and identifiability questions remain unresolved. This work is not offered as a validated predictive risk model.
Public research hubThe practice
Saudi Arabia · Remote international collaboration
Independent Researcher
& AI Technology Developer
My work focuses on making technical claims easier to inspect and automation easier to reason about. It brings together Python workflow development, agent-trace provenance and reproducibility, and permission-aware system design.
I work with defined requirements, approved inputs and clear delivery criteria. Public examples distinguish research software and local demonstrations from deployed customer systems.
Alharbi AI Research & Solutions is a provisional working brand for Abdullah Alharbi’s independent practice. It is not presented as a registered company or an OpenAI partnership.
04 / Start a conversation
Describe the outcome, systems and constraints. A short brief is enough to begin a feasibility and scope discussion.
albdrani36@gmail.comPlease keep credentials and confidential records out of an initial inquiry. Sensitive data requires an agreed processing environment.
عبدالله الحربي، باحث مستقل ومطوّر تقنيات ذكاء اصطناعي من المملكة العربية السعودية. للتعاون البحثي والمشاريع التقنية محددة النطاق.
This helper works in your browser. Review and edit the draft before opening your email app.
No form submission, analytics, browser storage or automatic email sending. Your text stays on this page until you copy it or choose to open it in your email app.