AI automation
Agents that do real work instead of demos: a bilingual solution architect that replaces a two-week consulting engagement, a publishing pipeline that runs itself, and a travel agent that watches fares while everyone sleeps.
The shift
Ten years of production software taught me where the real cost sits: not in writing features, but in the manual back-office work that surrounds them — triaging leads, answering the same twenty questions, assembling documents, checking prices, moving data between systems that refuse to talk.
That work is now automatable, and the engineering discipline matters more than the model. An agent that runs unsupervised in production needs queues, retries, error handling, structured outputs, secrets management and a human gate on anything irreversible — the same instincts that keep a mobile release from breaking a million phones.
Everything below runs on a self-hosted n8n stack I operate on a Linux VPS — not a demo notebook.
AI Solution Architect
A bilingual (English and Arabic) agent, live in production for PhazeRo. A visitor describes a business problem in plain language; the agent returns a structured transformation blueprint — diagnosis, solution architecture, implementation roadmap, team composition and ROI projection — in 30 to 60 seconds, where the equivalent is a two-week consulting engagement.
- Prompt architecture and a structured-output schema keep results consistent across telecom, logistics, healthcare, manufacturing, finance, oil and gas, and government use cases.
- The output feeds a lead-generation funnel: qualified brief, automatic CRM record, internal alert, booked consultation — marketing traffic converted to sales pipeline with no manual triage.
- It also repositions the company: from software house to consulting partner, demonstrated rather than claimed.
KDP publishing pipeline
A full Amazon KDP publishing line, independent project: niche and keyword research, content and cover generation, manuscript assembly, then listing creation, metadata, pricing and submission.
- Orchestrated in n8n with browser automation, file generation, queueing, retries and error handling.
- A human approval gate sits before the final publish step — the pipeline never ships a title on its own.
- Hours of manual work per title collapse into a single supervised run.
Travel agency agent
An operations agent for a travel business, covering three jobs a small team can't staff around the clock:
- Fare tracking — scheduled polling of pricing sources, alerting clients when a route drops below their target price.
- First-line support — booking and policy questions answered from the agency knowledge base, with a clean handoff to a human when confidence is low.
- Lead capture — qualification pushed into the CRM with automated follow-up sequences.
The confidence-based handoff is the part I care about most: an agent that knows when to stop is worth more than one that always answers.
The stack
- Automation — n8n self-hosted and cloud, multi-step agent workflows, LLM APIs (OpenAI, Anthropic), prompt engineering, structured outputs, RAG, MCP servers and clients.
- Integration — REST APIs, webhooks and event-driven pipelines, scheduling, queues, retries, browser and web automation, CRM and payment integrations.
- Backend & data — Node.js, TypeScript, Python, PostgreSQL and Supabase, Firebase, Google Workspace, Slack / Telegram / WhatsApp bots.
- Infrastructure — Linux VPS, Docker, Vercel, CI/CD, environment and secrets management, logging and monitoring.