Use Cases
Five shapes of the same architecture.
A token factory is one methodology. It adapts to the domain. Two of the use cases below are delivered engagements — aviation, multi-year, in production. The other three are expert approaches rooted in that work: how I would build them, the trade-offs named, the regulatory context where it matters.
01
Technical documentation
Structured extraction from long, text-heavy engineering documentation. Traceable answers that hold up to regulatory scrutiny.
Read the use case- Status
Delivered
- Provenance
Delivered at a Nordic aviation company. On record.
- What changes
Engineers stop treating retrieval as a tax; reviewer workload shifts to judgement, not lookup.
- Regulatory context
Civil aviation safety regime. Citation trails back to the source document.
02
Scanned-page tagging
OCR, classification, and taxonomy stewardship for working archives that arrive as pixels rather than as a schema.
Read the use case- Status
Delivered
- Provenance
Same aviation engagement as technical documentation.
- What changes
Tagging flows at the rate scans arrive; humans move from data entry to exception handling.
- Regulatory context
Record-keeping obligations under aviation maintenance regime.
03
Contracts
Clause-level retrieval, extraction, and comparison for organisations where the contract is the liability, not a convenience document.
Read the use case- Status
Expert approach
- Provenance
Expert approach based on adjacent delivered work.
- What changes
Legal hours shift toward the hard third; summaries stop being a legal-team bottleneck.
- Regulatory context
DORA Article 30 register of information, EU AI Act, sectoral rules for banks.
04
Knowledge retrieval
An internal knowledge layer over wikis, runbooks, specs, and policy libraries that you own end-to-end.
Read the use case- Status
Expert approach
- Provenance
Expert approach based on adjacent delivered work.
- What changes
Onboarding gets measurably shorter; tribal knowledge is surfaced instead of lost.
- Regulatory context
GDPR data-residency and retention controls as first-class parameters.
05
Decision support
Scoring, triage, and prioritisation with structured outputs, evidence pointers, and a replayable eval harness.
Read the use case- Status
Expert approach
- Provenance
Expert approach based on adjacent delivered work.
- What changes
Routine load drops; the hard cases get more — not less — human attention.
- Regulatory context
EU AI Act high-risk obligations: data governance, logging, human oversight.
Next step
Start with an architecture review.
Most conversations about a specific use case turn into an architecture review first. It is the cleanest way to separate what to build from what to buy.
Get in touch