I build AI systems that survive production.

Most AI projects die between the demo and the deploy. I've shipped systems that process millions of events a month in banking, real estate and enterprise customer operations, and I'll build you one that actually runs.

Book a 30-min call See what I've built

No pitch, no sales rep. You talk to the person who builds it.

19M+

Events a month processed by production systems I designed and shipped

7 yrs

Building production ML and data systems under real compliance and uptime pressure

$23M+

In profit generated by the systems I've designed and shipped

The gap

You've already tried AI. Your operations didn't change.

The blocker is rarely the model. It's that nobody built the thing to survive contact with real volume, real edge cases and real people.

Stalled

The pilot worked, then stopped

A demo impressed everyone in the room and never touched a real workflow.

Manual

Your team is the integration layer

People re-type the same data between four systems because nothing connects them.

Orphaned

The last automation broke

Built by someone who left, or a contractor who vanished. Now everyone works around it.

What I build

From one stubborn process to systems that run your operations

Workflow & process automation

Systems that move work between your tools without a person in the middle: routing leads, processing documents, syncing records, chasing follow-ups. The unglamorous work that quietly costs you a headcount.

AI agents that take action

Not chatbots that answer questions. Agents that read the email, extract the order, update the CRM and flag the exception for a human, with the judgment to know which is which.

Voice & conversational AI

Speech-to-text pipelines, call classification, emotion and intent detection, automated compliance checking. Built for operations handling hundreds of thousands of calls a month, not for a demo.

Detection & monitoring at scale

Real-time anomaly, fraud and exception detection on high-volume streams, where latency and false-positive rates decide whether a system is useful or just noisy.

How engagements work

Four steps, in this order

Audit

10 business days

I map your operations, interview the people doing the work and inventory your systems. You get a written report: every automatable process, ranked by hours and dollars recovered, with effort estimates. Yours to keep whether or not we go further, including if the honest answer is that automation isn't worth it here.

Design

1 week

We pick the highest-value target and I write the spec: what gets built, what it touches, where it runs, how it fails, what it costs to operate. Fixed scope before anyone writes code.

Build

2 to 6 weeks, scoped up front

I build it, on your infrastructure and your accounts. You see working software every week, not a status update.

Handover

Plus 30 days support

Documentation, a recorded walkthrough and the code in your repository. Your team can maintain it without me. Then 30 days of support included while it settles.

You own everything. The code, the accounts, the documentation, the vendor relationships. No platform lock-in and no proprietary layer you have to keep paying me for.

Selected work

Systems running in production

Designed and shipped before founding Unwirelab, in senior ML and data engineering roles. Built for large enterprises under real compliance and uptime pressure. I now bring the same engineering discipline to smaller teams, without the enterprise timeline.

Enterprise customer operations / Client unnamed under NDA

500,000 customer calls a month, turned into revenue and compliance signal

A multi-billion-dollar B2C service provider was generating half a million support and sales calls a month. All of it was effectively write-only: unsearchable, and feeding nothing downstream.

Customer emotion: at-risk accounts surfaced for retention intervention
Call classification: the entire archive searchable by reason for contact
Missed upsell detection: revenue opportunities that went unaddressed, flagged
Compliance verification: every call checked, instead of the 1 to 2% sample manual QA covers

Equivalent manual effort replaced

~800,000 analyst-hours a year

Moving compliance review from a ~2% sample to 100% coverage is equivalent to roughly $16M of manual review labor at typical loaded QA rates, on top of the retention and upsell signal the same pass produced.

The hard part: four inference passes over 500,000 calls a month at a cost per call that didn't exceed the value it created, and accurate enough that compliance output could be trusted in a regulated context.

Equivalent-effort estimate: 6M calls/year × 8 min/call manual review ≈ 800,000 hours, costed at $20/hour loaded. Illustrative of scale; not a client-reported figure.

Real estate & automotive / Al Habtoor

Time-to-lead down 95% across 25,000 leads a month

25,000 inbound leads a month were reaching sales agents slowly and unevenly. In a market where the first responder usually wins the deal, that delay was lost revenue.

Instant routing: to the right agent on arrival, by rules rather than by hand
Automated follow-up reminders: so leads stopped going quiet
Automatic re-assignment: when an agent didn't act in time, so no lead went unowned

Result

95% faster average time-to-lead

The hard part: the re-assignment logic. Too aggressive and agents claimed leads they weren't working just to hold them; too lenient and leads went quiet again. It had to change how agents behaved, not just where leads landed.

Banking security / OneSpan

Real-time fraud detection across 650,000 transactions a day

650,000 banking transactions a day needed screening for suspicious activity in real time, without burying the analyst team in false positives. I built a detection system that flagged suspicious activity blind, surfacing risk for review without exposing underlying account detail to the reviewing layer.

Scale screened

~$7.6M/year in expected fraud exposure

The transaction flow under screening carries roughly $7.6M a year of expected fraud exposure at published industry loss rates.

The hard part: the precision/recall tradeoff at volume. At 650,000 transactions a day, a false-positive rate that looks fine on a slide becomes thousands of wasted analyst-hours a week.

Estimated from published industry data: 650k transactions/day × $50 average transaction value × 6.4 bps global card fraud loss rate (Nilson Report, 2024). Illustrative of scale; not a client-reported figure.

Engagement model

What working together looks like

Start here

It starts with a paid audit

Fixed fee, 10 business days, and you keep the report regardless of what you decide next. It's the cheapest way to find out whether I'm useful to you, and the fastest way for me to tell you if I'm not.

Pricing

Fixed scope, fixed fee

We agree what gets built and what it costs before any code exists. No hourly billing, no scope drift, no surprise invoices. I'll give you a firm number after the audit, or a realistic range on our first call if you need one to plan around.

Timeline

Most projects run 2 to 6 weeks

Bigger systems get broken into pieces that each ship independently, so you're never waiting months to see something work.

Ownership

You own the output

Code in your repository, accounts in your name, documentation your team can act on. Nothing depends on me staying involved.

I'll tell you when to walk away. If the audit says a process isn't worth automating, that's what the report will say. I'd rather lose the build than sell you something that doesn't pay for itself.

The comparison

What changes

Without

With

Without Manual work quietly absorbs a headcount you never budgeted

With The process runs itself and your team does the work you hired them for

Without Tools bought, half-adopted and worked around

With Systems your team actually uses, because they were built around how they work

Without An automation nobody owns, breaking silently

With Documented systems your team can maintain and extend

Without AI pilots that impress and then stall

With Software in production, doing the job every day

Questions

Before you book

How do we communicate during a project?

Slack or your channel of choice, plus a weekly working demo, not a status report. You'll see the system doing something new every week. I'm responsive within the business day and I don't disappear between milestones.

Who owns what you build?

You do. Code in your repository, accounts in your name, documentation included. There's no Unwirelab platform layer and nothing you have to keep paying me for.

What happens if you're unavailable?

Every project ships with documentation and a walkthrough recording specifically so it doesn't depend on me. For systems that need guaranteed coverage, I'll say so up front rather than after you've signed.

Do we have to change tools?

No. I'm not a reseller and I don't have a preferred platform to push. I build on what you already use unless there's a concrete reason to change, and if there is, I'll show you the math.

How is our data handled?

Systems run on your infrastructure, so your data stays in your environment. I'll sign your NDA or DPA. For regulated workloads I've built under compliance constraints before, including EU data-residency requirements. See our privacy policy.

What does it cost?

The audit is a fixed fee. Builds are quoted fixed-scope after the audit, and I'll give you a realistic range on our first call so you can plan. I don't bill hourly.

What if the audit says don't automate?

Then that's what it says, and you keep the report. It's happened, and it's a better outcome than a build that doesn't pay for itself.

Find out what's automatable in 30 minutes

Bring one process that's costing you more than it should. I'll tell you whether it's worth automating, roughly what it would take, and what I'd do first. On the call, not in a proposal afterwards.

Book a 30-min call Price your manual work first

No pitch. No sales rep. Just an honest read.