From the Diary of Lorin Bejleri

Lorin Bejleri - Main Feature

Lorin Bejleri

From 30 Days Inside An Aircon Business To 50 Live AI Calls On One GPU: How Lorin Bejleri Builds For The Real Back Office

The Reality Of The Floor Over The Promise Of The Cloud

The decision to stop writing code and step inside a customer’s business for a full month is rare in software development. Lorin Bejleri did exactly that before building the second version of his platform. He watched the daily friction of service operations firsthand. He saw that artificial intelligence promises massive efficiency, but small business owners prioritize basic survival. They need systems that do not break when a customer reschedules at the last minute or a crew runs late. This ground-level truth dictated a radical technical approach, proving that the best software is shaped against real operations rather than boardroom assumptions.

Engineering Trust For The Blue-Collar Backbone

Lorin Bejleri is the Founder and CEO of CrewRun.AI, where he builds autonomous agents that run the back office for everyday service businesses. He writes code, sells what he builds, and presents his technical architecture to rooms that matter across the globe. His defining professional trait is an absolute refusal to let technology outpace operational reliability.

Fifteen Years Of Selling Complex Systems To Skeptical Buyers

The path to building autonomous business systems started long before the current technological wave. Lorin spent fifteen years building software and eight years selling complex artificial intelligence to non-technical buyers. His early work involved designing computer vision and deep learning solutions as a Senior Manager at Xjera Labs. He deployed these advanced systems for Singapore government agencies and international clients.

He learned quickly that technical elegance means nothing if the buyer does not trust the output. This realization shaped his independent services practice, MediaDua. He served Singapore SMEs for seven years and saw the administrative burden crushing local business owners.

Later, as the fractional Chief Technology Officer for FutureGrail, he architected the technology stack for a rapidly growing online auction house. He also founded SurfRank to solve visibility analytics for companies lost in algorithmic recommendations. Every role reinforced a single truth. Business owners do not care about the underlying technology. They care about whether the system makes them faster, richer, or less stressed.

Computing Prices Deterministically And Keeping Humans In Control

Today, Lorin focuses entirely on building CrewRun into the definitive booking layer for service businesses. The goal is to take a customer message and turn it into a paid invoice autonomously. He started by addressing the hardest bottleneck in operational technology.

“Trust, specifically accuracy with accountability,” Lorin says. “An owner will let AI answer a customer, but won’t let it send an invoice or book a job until it’s proven it won’t make a costly mistake.”

This understanding drives the entire product architecture. The system drafts quotes using the business’s actual rate card. It computes prices deterministically rather than relying on generative models to guess a number.

“My advice: don’t make the model do what it’s bad at,” he explains. “We compute prices deterministically from real data, never letting the LLM do math.”

By keeping the language model away from calculations, CrewRun prevents costly hallucinations. A customer message lands in the inbox. The system drafts a quote, recommends a crew, and waits for a single tap of approval from the owner. Once approved, it books the job, dispatches the team, and builds a service report from on-site photos. Finally, it issues the invoice and chases the payment.

To make this financially and operationally viable, Lorin made a massive architectural decision. He moved away from expensive third-party APIs and built a fully self-hosted stack.

“We built our own voice-to-voice agent that responds in under 1.2 seconds, running roughly 50 concurrent conversations on a single GPU,” Lorin notes. “The impact is 85 percent gross margins, no per-call API costs, and customer data stays on our own infrastructure.”

The speed is critical for user adoption. A sub-800 millisecond delay to the first audio makes the voice agent feel entirely natural. It is the difference between a neat demonstration and a tool an owner relies on daily. The results of this technical precision are measurable and stark.

“Our design partner cut admin from about three hours a day to thirty minutes,” he says. “He now runs his entire business on it solo.”

Lorin Bejleri achieved this by staying close to the end user. He protects his engineering cycles by shipping in tight loops against actual customer behavior. Building, selling, and presenting are not separate jobs for him. The same insight that closes a customer drives the next build. The system measures approval-override rates closely to see where the software still needs human oversight and where it has earned full autonomy.

The Final Measure Of Operational Software

The month spent watching a service business operate yielded a system that actually works in the field. Lorin Bejleri did not build a wrapper around an expensive interface. He built a self-hosted, lightning-fast engine that understands the fragile trust between a business owner and their bottom line. The true test of back-office technology is not how smart the code appears. It is whether the owner can finally put their phone down and go home.

Lorin Bejleri is the Founder and CEO of CrewRun.AI based in Singapore. He builds autonomous agents that run the back office for everyday service businesses. To connect with Lorin or learn more, visit his LinkedIn profile.

Share post:

Subscribe

Popular

More like this
Related

This website is for preview purposes only. The stories here are available as a preview exclusively for our fellow Executives Diary members before they are published on the main website. These blog posts are not indexed by Google, as we have restricted search engine access to this preview site.