How I Built 4 AI Products While Working Full-Time at a Bank

I'm a Senior Retail Banking Officer at Emirates NBD — one of the largest banks in the Middle East. Full shifts. Client relationships. Sales targets. GEMS Sapphire Award (top 1% performer).

I also built 4 production AI products in the past year. One of them has 1,600+ source files.

People ask "how." Usually they expect some inspirational productivity hack. The real answer is more boring and more useful.

The Honest Time Budget

My Actual Week (averaged over 6 months)

Monday–Friday:

• Bank work: 8–9 hours/day
• Commute: 1 hour/day
• Evening building: 1.5–2.5 hours (not every day)
• Available evenings: ~3–4 per week

Weekends:

• Saturday: 4–6 hours (primary build day)
• Sunday: 2–4 hours (lighter, planning + writing)
• Sometimes: full Sunday sprint (rare)

780 hours. That's what produced 4 AI products. Let me break down where those hours actually went.

What 780 Hours Produces

Hour Allocation (approximate)

Premier Radar took the lion's share because it's an enterprise-scale platform. The others were simpler — but they exist, they work, and they prove I can deliver across different architectures and problem spaces.

The System That Actually Works

Rule 1: One Project at a Time

I never context-switch between projects within the same week. When I'm building Premier Radar, everything else is frozen.

Why this matters — context-switching cost for AI projects:

With 15 hours/week, losing 60 min to context-switching means losing 7% of your total available time. Twice a week = 14%. That's an entire evening session gone.

Rule 2: Session-Based Development

Every coding session follows a strict protocol:

Session Protocol

Start (5 min):

• Read last session's handoff note
• Identify ONE concrete deliverable for today
• Time-box the session (usually 2–3 hours)

Build (2–3 hours):

• Focus on the single deliverable
• No "quick improvements" to other things
• No refactoring unless it's the stated goal
• Ship or reach clear stopping point

End (5 min):

• Write handoff note for future me:
  • What I finished
  • What's next (concrete, not vague)
  • Any bugs or gotchas discovered
  • Architecture decisions made + reasoning
• Commit everything, push to GitHub

This handoff system is the most important productivity technique I've found. Without it, I'd lose the first 30-60 minutes of every session re-loading context. With it, I'm productive in 5 minutes.

Rule 3: Ship Ugly, Iterate Fast

My Shipping Philosophy

What never ships: The version that's "almost perfect" but has been sitting on your laptop for 3 months.

The AI Leads Portal's first version had hardcoded product categories and no authentication. But it worked at a live bank expo event, and it proved the concept. That's worth more than a perfect app that nobody's seen.

Rule 4: Use Your Day Job as Input

This is the unfair advantage that most side-project builders don't have.

Every day at Emirates NBD, I sit across from the exact users my AI products serve. I watch relationship managers handle leads. I see the compliance workflows. I know which processes are painful because I live them.

What most side-project builders do:

1. Guess
2. Build
3. Hope someone wants it

What I do:

1. See real problem at work
2. Validate it's widespread
3. Build on weekend
4. Test with real workflow
5. Iterate

This is why my products work. They're not solutions looking for problems. They're problems I've lived with for years, finally getting solved.

The Mistakes That Cost Me Weeks

I want to be honest about what didn't work:

Trying to build every evening: Burnout in 2 weeks. Now I build 3-4 evenings per week max. The other evenings are for family, rest, and thinking.

Feature creep on Premier Radar: I spent 3 weeks building an analytics dashboard that nobody asked for. Those were 3 weeks I could have spent on the core AI scoring engine. Lesson: build what's needed, not what's cool.

Not writing tests early: Month 3, I broke the lead scoring algorithm with a "small change." Took 2 days to find the bug. Now I write tests for anything that touches scoring or financial calculations.

Comparing myself to full-time builders: Twitter is full of people shipping daily updates because building IS their job. I have 15 hours/week. Different game, different strategy. Once I accepted that, I stopped feeling behind and started being strategic.

The Results

12 Months, 15 Hours/Week:

Total: 4 products in production, all built alongside a full-time senior banking role with top 1% performance rating.

You don't need to quit your job to build serious AI products. You need a system that respects your constraints, tools that multiply your limited time, and the discipline to ship imperfect things consistently.

The products aren't perfect. But they exist, they work, and they prove capability at a level most full-time builders haven't reached.

That's the point.