Karam Hawash is the founder of GenovaFlow, an AI automation agency based in Dubai. GenovaFlow builds AI agents that run real work inside a client's existing tools: voice agents that answer calls, chat agents on WhatsApp and the web, outreach agents that follow up on leads, and custom agentic systems for back office work.
Every week Karam takes one tool or technique, usually Claude Code, n8n, an MCP server or a new model, runs it on his own projects and publishes the result on his YouTube channel: the setup, the numbers on the screen, the cost and the parts that failed. The articles on this blog are written from those tests and from client builds, and the files from each video are free in his community.
He writes for founders, operators and builders who want to know whether an AI tool will actually save them time before they spend a month on it.
Posts by Karam Hawash
11 Fixes That Cut My Claude Code Token Usage by 70 Percent
Two installs, rtk and caveman, and nine built in Claude Code habits took my session from 100 percent used to 20. Each fix, its command and why it saves tokens.
Read the post
When should an AI agent hand off to a human?
Hand off when the agent keeps failing, when an action is risky or cannot be undone, or when the customer asks. What OpenAI, Anthropic and vendors say.
Read the post
What "plugs into your stack" means for AI agents
APIs, webhooks, OAuth and MCP are how an AI agent reads, acts and writes back in your CRM, calendar and helpdesk. What each one does and what to ask.
Read the post
How multilingual WhatsApp agents capture after hours bookings
A WhatsApp agent can answer a 1am booking request in the guest's own language. How Meta's 24 hour window, templates and pricing shape that agent.
Read the post
AI voice agent vs a human receptionist: the cost math
A US receptionist costs about $55,330 a year with benefits; a voice agent bills $0.14 to $0.17 a minute. The math, with every input linked to its source.
Read the post
Turn one YouTube video into LinkedIn, X and Reddit posts with n8n
An n8n workflow that turns a YouTube video into LinkedIn, X, Reddit and blog posts, with a Telegram approval step before anything publishes.
Read the post
Host n8n on the Google Cloud Free Tier with Docker
Run n8n in Docker on Google Cloud's free e2-micro VM with Nginx and a free Let's Encrypt certificate. Every command checked against the official docs.
Read the post
Brain chips and BCIs: what has happened and what is speculative
Brain computer interfaces today: peer reviewed speech results, Neuralink and Synchron trial milestones with dates, and the claims that are still plans.
Read the post
Build a Gemini Voice Agent for Telegram in n8n
An n8n Telegram bot that transcribes voice notes with Gemini, answers with Gmail and Sheets tools and replies by voice, updated for current Gemini models.
Read the post
How large language models differ, and what to compare
Context window, reasoning effort, tool use, input types and price per token: how LLMs from Anthropic, OpenAI and Google differ, from their own pages.
Read the post
Where AI is heading: what the data shows and what is forecast
Where AI is heading, from dated reports: Stanford's AI Index, Epoch AI, METR and the IEA. What has been measured, what is forecast, and who said it.
Read the post
AI in robots: what is deployed today and what is still a demo
What AI robots actually do today, from IFR counts, Amazon, Waymo and dated humanoid pilots, which claims have evidence, and the limits the industry names.
Read the post
How a neural network learns, from weights to backpropagation
A neural network learns by measuring its error and adjusting its weights with gradient descent and backpropagation. The layers, the loss and the 1986 method.
Read the post
n8n vs Zapier vs Make for AI workflows, and where Claude Code fits
How n8n, Zapier and Make price AI workflows, what n8n's license lets you self host, and when a coded agent like Claude Code is the better tool.
Read the post
RAG explained: how retrieval works and when to use it
Retrieval augmented generation finds the right passages and puts them in the prompt. How it works, from the 2020 paper to vector search, and when to skip it.
Read the post
What agentic AI is and how it differs from a chatbot
Agentic AI is a model that picks its own next step and uses tools in a loop. How it differs from a chatbot or a workflow, with a worked example.
Read the post
Companies Using AI in Every Task: What the Surveys Show
How companies actually adopt AI across their work, from Stanford's AI Index, US Census and Eurostat surveys: which tasks come first and what gains firms report.
Read the post
AI Agents in Real Estate: Lead Response, Viewings, Listings
Where AI agents help a brokerage today: answering leads, booking viewings and drafting listings, plus the consent, fair housing and Dubai permit rules.
Read the post
AI Agents in Financial Services: Onboarding, Support, Rules
Where AI agents work in financial services today (onboarding, document checks, client support), what FCA, ESMA, FINRA and the EU AI Act expect of them.
Read the post
AI Agents in Healthcare: Intake, Booking and HIPAA Rules
Where AI agents do real work in a clinic today (calls, booking, reminders, intake), what HIPAA requires of the vendor, and six checks before going live.
Read the post