What we source
Lettus AI works with operating companies to source and prepare material they own, produced in the course of running their business. The supply spans written and structured records as well as media captured where the work happens.
- Workflow and process records — SOPs, playbooks, tickets, and project history
- Operational documents — docs, slides, sheets, PDFs, and email threads
- Field and on-site video — task capture, walkthroughs, and inspections
- Real-world audio — calls, dispatch, shop-floor and field recordings
- Images and visual records — photos, scans, diagrams, and annotated captures
- Expert feedback and review — corrections, QA judgements, and annotation from qualified practitioners
Sourcing is organised by domain and workflow rather than by file type, so a dataset can combine several of these formats around the same task.
Original material and processed datasets
We distinguish between the original material a company contributes and the processed dataset prepared from it. Original files — including raw video, audio, and images — are sourced into a controlled processing environment scoped to the agreed use case.
What a partnership licenses is the processed dataset: records that have passed through de-identification and preparation steps agreed with the contributing company before any delivery.
Rights and provenance
Every contributing company is asked at intake about ownership of the material, whether identifiable people appear in it, and whether consents or releases are in place. Answers that indicate mixed or unclear rights are reviewed before a dataset moves forward.
Provenance is recorded per dataset: which company contributed the material, what business activity produced it, and what the company represented about its rights to contribute it.
Privacy review
Text and structured records pass through the de-identification pipeline described on our data security page: direct identifiers are redacted, tokenised, or removed, and free-text fields are scrubbed for entities that carry personal information.
Media carries a different privacy surface. Faces, voices, name badges, screens, licence plates, and location detail can all appear incidentally in footage recorded in a working environment, and they are identified during review rather than assumed absent.
Licensing
Each dataset is licensed under terms agreed with the contributing company. Companies retain ownership of their underlying material; a partnership grants defined, scoped permissions rather than transferring ownership.
Permitted use, retention, and deletion are written into the agreement before any dataset is prepared, and the contributing company reviews representative samples before prepared data is used.
How an engagement starts
Tell us the capability you are trying to improve and the shape of data that would move it — domain, formats, volume, and refresh cadence. We work backwards from that to identify which operating companies could contribute, and what preparation the material would need.
We do not maintain an off-the-shelf catalogue. Sourcing is commissioned against a defined requirement, and scope, rights, privacy review, and licensing terms are settled before collection or preparation begins.