Agentic AI / LLM systems / rapid prototyping
01 / Selected work
Every project starts with a practical problem and a real user. The work below shows how I deliver AI systems: interview the people involved, build the MVP fast, then validate it with them until it holds up.
02 / Built at Omantel
Since moving into the Agentic AI Engineer role in September 2026, I have architected and built three LLM tools with AI coding agents, each shaped through stakeholder interviews and validated with the teams who use them. The data is internal, so the work is described rather than demonstrated.
Database operations / LLM diagnostics
A FastAPI and Claude tool for the DBA team. It reads AWR reports, alert logs and trace files, masks sensitive identifiers before anything reaches the model, and returns a root-cause report with IF/THEN diagnostic commands, an evidence timeline and a follow-up agent for questions.
Service desk / bilingual retrieval
A pop-up assistant on Omantel's ticketing system that lists a group's open tickets and retrieves the matching knowledge-base documents in Arabic and English. Built as a light read-only overlay so it survives a future platform upgrade.
Business intelligence / prototype agent
A prototype agent that learns the BI team's conventions from existing Power BI dashboard scripts to automate data ingestion, data reading and dashboard generation.
Archived prototype / ranked 1st
An LLM-powered assistant built to read an entire email thread, extract relevant context, and draft a response with configurable controls. The prototype was archived after Omantel adopted Microsoft Copilot.
Ranked 1st in Omantel's Gen Z AI project competition.
03 / Independent builds
A bilingual Arabic/English AI phone receptionist for Gulf small businesses, built on Pipecat, FastAPI, Postgres and Cal.com. Real-time browser voice demo complete; Twilio phone integration in progress.
View repositoryTurns any public GitHub repository into a 3D city with TypeScript and Three.js, with Cloudflare Workers ingestion, caching and rate limiting. CI runs 285 tests, type checks and deployment dry runs.
View repositoryAn agent pipeline on a VPS that gathers, validates, ranks and delivers a daily AI news briefing to about 10 readers, with cross-referencing and archive deduplication built in.
View repository04 / Practice
Agentic workflows, tool use and MCP, RAG and document retrieval, output validation, sensitive-data masking, prompt engineering, and voice agents.
Anthropic Claude API, open-weight and locally hosted LLMs, LangChain, LangGraph, and Pipecat.
Python, FastAPI, TypeScript, Docker, Cloudflare Workers, GitHub Actions, pytest, and Vitest.
SQL, Greenplum, Power BI, data warehouse migration, and n8n.
05 / Profile
Omantel, AI Lab
Omantel, Data Analytics & AI Platforms
Omantel, Data Science & Machine Learning Programme with Codeline
Jarzez Trading LLC, operations and project coordination
University of Reading, United Kingdom
Languages Arabic (Native) / English (Fluent)