I build automation systems that run without you, and know when to hand a decision back to a human.

The challenge

A Dubai real estate company had 20-30k leads, a mix of old backlog and new incoming, and no way to qualify them before they reached sales. Two problems hid in one project: bulk qualification that wouldn't get their WhatsApp number flagged by Meta, and instant handling of fresh leads. Most builds get this wrong by treating them the same.

The system

• A Claude-powered agent ("Sarah") holds a natural WhatsApp conversation and extracts five qualification fields from freeform replies: budget, buy vs. rent, financing status, preferred area, and timeline.
• Each answer feeds a weighted scoring engine. Leads are tagged Hot, Warm, or Cold automatically, using thresholds the client signed off on, not ones the AI guessed.
• Hot leads create a deal in the pipeline with a structured summary note, so an agent knows everything in 10 seconds.
• Cold leads drop into a nurture sequence with timed re-engagement.
• Backlog handled with Meta-approved templates in throttled batches to protect number quality; new leads trigger instantly via webhook.

Outcomes

Stack

The challenge

Mind, Behavior and Psychiatric Clinic needed steady patient acquisition content, but producing videos manually in a compliance-sensitive field was slow and inconsistent. As their Growth Strategist, I own both the marketing strategy and the system that executes it.

The system

• A 34-node n8n workflow that takes a content calendar from idea to published video with no manual editing steps.
• Claude AI writes the scripts, ElevenLabs generates voiceover, HeyGen produces the video, and Cloudinary hosts the media.
• Compliance-sensitive steps stay behind human review before anything goes live. Routine production steps run on their own.
• Publishing is automated across 4+ platforms, alongside conversion funnels and compliance-first paid campaigns targeting mental-health service seekers.

Outcomes

Stack

The challenge

Running a consistent YouTube Shorts channel normally takes 15 to 20 hours a week of research, scripting, editing, publishing, and analytics review. The target for this build: keep the channel publishing daily on under three hours of human time per week, without lowering editorial standards.

The system, in five stages

• Ideation: five agriculture news feeds merged and analyzed; Claude proposes ranked video ideas in a "farm-business mentor" voice.
• Scripting: every script must deliver both practical how-to and business ROI. That dual-angle rule is enforced in the prompt itself, and the AI is barred from inventing figures.
• Production: scripts rendered into Shorts via JSON2Video with stock footage matched from Pexels.
• Publishing: scheduled uploads via the YouTube Data API, metadata prepared automatically.
• Learning loop: weekly analytics feed back into ideation, so the pipeline gets smarter about what performs.
• Email approval gates sit between every stage. One tap from a phone approves the next step.

Outcomes

Stack

Productized

This pipeline is being packaged as a done-for-you service for other channel owners. If you run a faceless channel and want your publishing handled end to end, get in touch.

The challenge

A psychology content brand needed a steady stream of quote graphics for social feeds. Producing them manually meant writing a quote, briefing a designer, waiting for the visual, and posting it. That cycle collapsed whenever the team got busy, and consistency is the whole game in content.

The system

• A schedule trigger starts each run and pulls the next title idea from a Google Sheets queue.
• A Quote Generator agent writes the psychology fact and caption copy, then logs it to the sheet.
• A second Design Generator agent turns that copy into a visual brief, generates the image, and renders the final graphic from a locked brand template so every post stays on-brand.
• Finished assets are uploaded, and the sheet is updated with the image record, giving the team a full archive of what was produced and when.

Outcomes

Stack