Dubai · CRM, AI automation, agent systems

I build AI systems that do the repetitive half of the work, with a person still in charge.

Six years on the customer side showed me where teams lose hours. Now I design the workflows, agents and quality gates that give them back.

See the systems
Claude
Classify
delivery · not_collected
priority
Gate
Quality gate
≥0.90 · PII · dup
confPIIdup
HubSpot
Ticket #18422owner: Ali
Slack
#ops-delivery
posted with context
Branch
Human review
below the gate
09:415G ▮▮▮
+971 5• ••• 4821WhatsApp Business
Today
my order 4821 was not collected, I'm waiting at the door09:41
deliverynot_collectedpriority
RULE-CHECKED
Sorry about that. Ali from our delivery team is on it, pickup slot in 10 minutes09:42 ✓✓
Message
  1. 1
    Classify (Claude)delivery · not_collected · priority
  2. 2
    Quality gate≥0.90 · PII · dup, else Human review
  3. 3
    HubSpot ticket #18422Slack #ops-delivery · owner: Ali
2h → 20mmanual handling / agent / day
100–200messages routed daily
+21%campaign engagement
95%VIP retention
−25%escalation time
What I fix

Problems I hear from CX and CRM leads.

Pick one. You get the symptom in their words, what I build, where the gate sits, and the number after.

01 / 06
Who feels itHead of Support

“Our agents spend the first two hours sorting and tagging before they talk to anyone.”

What I build
  1. Claude reads each Slack, WhatsApp and email message and labels intent, priority and team.
  2. A ticket is created in HubSpot with an owner and a deadline, and the team channel gets the context.
  3. Agents open their day to a sorted queue instead of a raw inbox.
The gate
Anything under a 0.90 confidence score goes to a short human review queue.
2 h → 20 min
manual handling per agent per day
  • Claude
  • HubSpot
  • Slack
  • n8n

Who feels itHead of Support

“Our agents spend the first two hours sorting and tagging before they talk to anyone.”

What I build
  1. Claude reads each Slack, WhatsApp and email message and labels intent, priority and team.
  2. A ticket is created in HubSpot with an owner and a deadline, and the team channel gets the context.
  3. Agents open their day to a sorted queue instead of a raw inbox.
The gate
Anything under a 0.90 confidence score goes to a short human review queue.
2 h → 20 min
manual handling per agent per day
  • Claude
  • HubSpot
  • Slack
  • n8n

Work

Systems in production, with the number after.

Each one started as a manual loop and ended with a gate, an owner and a measurement.

Support operations01 / 03

Dispatch

2 h → 20 min
2 h20 minbeforeafter
Problem
Agents sorted and tagged by hand every morning.
Built
Classification and routing for every Slack, WhatsApp and email message; tickets created in HubSpot with an owner; a 90% confidence gate sends the rest to a human queue.
Result
2 h → 20 min per agent per day; 100–200 messages a day.
  • Claude
  • HubSpot
  • Slack
  • n8n
Lifecycle marketing02 / 03

Win-back

+21% engagement
EmailWhatsAppPush
Problem
Campaigns went out late and to the wrong people.
Built
Segments by churn risk, win-back and VIP journeys, copy for Email, WhatsApp and Push, rule-checked before scheduling in HubSpot.
Result
+21% engagement; 95% VIP retention.
  • HubSpot
  • Zapier
  • WhatsApp
Escalations03 / 03

Ownership

−25% resolution
Escalation inOwner + deadlineDaily QA · 20 sampledResolved
Problem
Escalations had no owner and no clock.
Built
Routing with ownership and deadlines, a daily QA sample of 20 routed items, dashboards for leadership on resolution and retention.
Result
−25% resolution time; 89% CSAT across 60 tickets a day.
  • HubSpot dashboards
  • QA
How I work

From the manual loop to the number in writing.

You see the manual loop before I touch it, and the result after.

  1. Find the loop

    A week shadowing the team and the inbox. Where the hours go, in minutes, per person.

  2. Build the workflow

    n8n or Zapier for the plumbing, Claude for the judgement, HubSpot and Slack as the destinations.

  3. Put a gate on it

    Confidence threshold, PII and duplicate checks, a human queue for the rest, a daily QA sample.

  4. Measure and hand over

    A dashboard the team owns, a runbook, and the before/after number in writing.

About

Who is Rukhillo Kodirov?

I'm Rukhillo Kodirov, a CRM and AI automation specialist based in Dubai, United Arab Emirates. I have spent six years in customer experience, CRM and operations, the last two building AI automation. Today I build CRM and AI automation at a UAE on-demand services company.

I build the workflows that classify, gate and route customer messages in HubSpot, n8n, Zapier, Slack and WhatsApp, with a human in the loop. In my current role they cut agent handling time from 2 hours to 20 minutes a day. I work in English, Russian, Persian, Turkish and Korean.

Read the full profile →See the nine certificates →
Stack

The tools I work in.

Have a team losing hours to manual work?

Send me the problem in two lines. I'll reply with what I'd build and what it would change.