Key Takeaways

  • Aerocore Technologies runs engine wash shifts from 7:00 PM to 5:00 AM across airport tarmacs, where solo technicians enter work data directly into mobile phones under poor conditions.
  • Manual data entry during overnight shifts creates recurring errors, particularly inverted or fat-fingered aircraft tail numbers that corrupt billing records.
  • Aerocore automated quality control with a computer vision agent that scans every work photo at 8:15 AM each morning, comparing images against work order entries.
  • The system routes discrepancies directly to management for human review instead of allowing the model to silently overwrite billing records.

The Midnight Fat-Finger Problem

Commercial aviation maintenance has a persistent field data problem: the people doing the physical labor work in the dark, and they work alone.

At Aerocore Technologies, crews service aircraft engines parked at airport gates and maintenance hangars worldwide. Darren Taillie points out the working conditions that destroy data quality: “So guys are out there on every plane doing a whole ton of work, including our team who works shifts from about 7:00 p.m. till about 5:00 a.m. doing these engine washes out on an aircraft.” (play from 39:16)

A single technician handles the entire wash procedure on a given engine. Once the job wraps up, that technician pulls out a smartphone to submit job metadata and proof-of-work photos. As Taillie notes, “There's no one out there to validate his inputs, his or her inputs, sorry. So, errors happen. Fat fingering happens.” (play from 39:53)

A technician finishing a ten-hour shift at 4:30 AM might swap two digits on a tail number. That small typo cascades: the airline gets billed for an aircraft parked three states away, the actual airframe shows an overdue wash in maintenance logs, and administrative staff spend days untangling work orders.

The 8:15 AM Vision Audit

Most software teams try to solve field data entry by making mobile forms stricter. They add validation rules, confirmation dialogs, or mandatory field reviews. Those measures fail on an active tarmac in freezing weather because they slow down workers whose primary job is mechanical maintenance, not data entry.

Aerocore bypassed the field form entirely. Instead of policing technicians during their shifts, Taillie built a batch verification pipeline powered by computer vision. Field staff take standard photos of their work and go home at 5:00 AM. Then the software steps in:

“Every day at 8:15 in the morning, agent goes in, scans all the pictures, verifies it against the information put in into the app.” (play from 42:06)

The vision model inspects the physical tail number, engine markings, and work order metadata visible in the photos. It compares those visual tokens directly to the text strings entered by the technician hours earlier. If the data matches, the work order clears for customer invoicing. If anything looks off, the pipeline stops:

“If it doesn't, it returns a flag. That flag gets sent to me and says, 'Hey, something's wrong with this work order. Please go have somebody, a human, go look at it and see what's going on here.'” (play from 42:39)

Taillie receives a daily digest outlining every completed job alongside each flagged discrepancy. A supervisor can review the image, correct the transposed character in twenty seconds, and clear the invoice before the morning accounting cycle runs.

Where This Method Breaks Down

Batch vision verification works well for retrospective billing audits, but it fails if your operation requires real-time safety interventions. Because Aerocore runs the agent at 8:15 AM, three hours after crews leave the field, the system cannot verify whether a physical step was missed while the aircraft is still accessible. If the plane pushed back at 6:00 AM, the photo audit only catches administrative discrepancies, not operational defects.

It also relies on photo quality from tired workers. If a technician submits an image with extreme lens glare from airport floodlights, motion blur, or bad framing that cuts off the registration number, the model cannot extract the text. That still generates a flag for human review, but repeated false alarms create alert fatigue for the team managing the morning digest.

What to Do With This

Audit your operations for single-operator field logs that trigger downstream accounting corrections. If field workers currently type identifiers (serial numbers, VINs, shelf tags) that already appear inside their required job photos, turn off manual validation checks on their phones. Schedule a morning batch script to run those photos through an optical character recognition model, check the extracted text against the database entries, and email your operations lead a single discrepancy list before daily standup.