Custom model for tech log defects
Hypothetical. Not a client, not a result.
// The problem
An airline's reliability engineers read a steady stream of free-text tech log entries to code them by ATA chapter and spot recurring faults, and the backlog never clears.
// What we would build
A custom language model fine-tuned on your own tech log history that codes each entry by ATA chapter, links it to earlier entries on the same aircraft and system, and flags possible repeat defects for an engineer to review. A custom model fits because tech log shorthand is full of abbreviations a general model misreads; it runs on a GPU server in your own data centre, so the records stay with you.
// What is in it
- Your tech log history cleaned and used to fine-tune the model
- Each new entry coded by ATA chapter, with the model's confidence
- Possible repeat defects linked across entries for the same aircraft and system
- Engineers confirm or correct, and corrections join the next training round
- Measured against a held-out set coded by your engineers before each release
- Runs on a GPU server in your own data centre
// Stack
- Fine-tuned open-weight LLM
- PyTorch
- vLLM
- Python
- Next.js
- Lenovo GPU server on site
// estimate
- Build
- ≈ US$171,000 to US$261,000, delivered within 33 weeksCAD 243,900 to 371,100
- Hosting
- ≈ US$176 a monthCAD 250 a month
- Support
- ≈ US$2,500 a monthCAD 3,565 a month
- Hardware
- ≈ US$47,800 to US$197,000CAD 68,000 to 280,800
Prices in your currency are estimates from today's Bank of Canada rate. All invoicing is in CAD or USD.
- AI route
- Custom LLM
- Model running cost
- ≈ US$176 a monthCAD 250 a month
Timeline by milestone
- Discovery
- 2.9 to 3.4 weeks
- Data and evaluation plan
- 0.4 weeks
- Design approved
- 2.4 to 4.4 weeks
- Core features
- 1.9 to 3.4 weeks
- Full build
- 1.4 to 2.4 weeks
- Training
- 7.4 to 12.4 weeks
- Testing and fixes
- 0.4 weeks
- Evaluation and red-teaming
- 3.9 to 4.4 weeks
- Launch
- 0.9 to 1.4 weeks
- Production
- 0.9 to 1.4 weeks
Outside our hands, and added to the calendar
- Hardware delivery, after it is ordered
- 2 to 6 weeks
- Test access to your maintenance system
- 1 to 4 weeks
How it is paid
- Deposit 20%
- CAD 48,540 to 73,980
- Discovery 1.5%
- CAD 3,640.50 to 5,548.50
- Data and evaluation plan 7.9%
- CAD 19,173.30 to 29,222.10
- Design approved 1.5%
- CAD 3,640.50 to 5,548.50
- Core features 4.6%
- CAD 11,164.20 to 17,015.40
- Full build 3.1%
- CAD 7,523.70 to 11,466.90
- Training 23.9%
- CAD 58,005.30 to 88,406.10
- Testing and fixes 1.5%
- CAD 3,640.50 to 5,548.50
- Evaluation and red-teaming 15.9%
- CAD 38,589.30 to 58,814.10
- Launch 1.5%
- CAD 3,640.50 to 5,548.50
- Production 8.6%
- CAD 20,872.20 to 31,811.40
- Holdback, 30 days after launch (10%)
- CAD 24,270 to 36,990
- GPU time for training and testing, at cost
- CAD 1,200. Billed up front, at cost, outside the milestones
- GPU server for AI, for the model
- CAD 68,000 to 280,800. Billed up front, at cost, outside the milestones