From voice to the 231 Model

Building an Organisation, Management and Control Model under Italian Legislative Decree 231/2001 is one of the most document-intensive processes in consulting: interviews with function heads, analysis of company records, organisation charts and procedures, mapping of processes against predicate offences, gap analysis, and finally the drafting of hundreds of pages of deliverables. Lekton Dossier is the proprietary software we designed and developed entirely at Lekton.AI to automate the whole chain, leaving the consultant what really counts: professional judgement. It is an already-working product, in use on real compliance engagements.

The workflow

It all starts from the interview recordings. The system extracts the audio, transcribes it with speech recognition and automatic speaker diarisation, and corrects the transcript with a language model that preserves proper names and sector-specific technical terminology thanks to protected vocabularies configurable for each engagement.

On top of the transcript, and of the acquired company documents (company record, organisation chart, current model, policies), sits the multi-pass 231 analysis pipeline we have developed: a sequence of specialised steps in which each stage produces a verifiable artefact — the narrative report, the structured process/risk matrix, the reconciliation between the two, the gap analysis across the Model’s 19 risk areas and the final risk scoring. The separation into distinct passes is not a technical detail: it is what makes it possible to validate every piece of evidence against what was actually stated in the interviews, with the ironclad “zero invention” rule.

The output deliverables are the ones the consultant would hand to the client: the risk matrix in Excel compliant with the multi-sheet professional template, the General Part of the 231 Model, the Code of Ethics and the action plan of the protocols to be drafted, all with the client’s branding. Every output is created in a staging state: nothing is considered final without the professional’s review.

Minutes for the Supervisory Body, Board and Statutory Auditors

The same technology covers a second use case that we are increasingly asked for: the automatic minuting of corporate body meetings. From the recording of a session of the Supervisory Body, a Board of Directors or a Board of Statutory Auditors, the system produces a complete formal record: constitution and quorum, agenda, treatment of the items in appropriate legal language, decisions, action items with owners and deadlines, and precise regulatory references.

Moving from speech to minutes is not a mere summary: each domain has its own register. Minutes of a Supervisory Body use the technical lexicon of 231 oversight — information flows, predicate offences, prevention protocols — whereas Board minutes follow their own conventions. The extraction is performed on a validated data schema that we defined and implemented in-house, which guarantees structurally complete and consistent minutes, rendered in Word with the firm’s formal layout. This pipeline is proprietary Lekton.AI technology: we are happy to show it in private demos.

Under the hood (for developers)

Lekton Dossier is a desktop application in Python with a PyQt6 interface, declarative pipelines configured in YAML with idempotent caching on inputs, data schemas validated with Pydantic, transcription via faster-whisper + pyannote.audio, a FAISS vector index for chat over the engagement, and native generation of the Word and Excel deliverables. It is the result of merging two codebases we developed and evolved on real projects, now consolidated into a single framework. Two basic components shared with our document platform are open source on our GitHub: ocr-cascade (multi-engine OCR with automatic fallback) and verified-citations (verification of AI citations against sources).

The consultant at the centre

Two design choices sum up our philosophy. The first: the prompts that drive the analyses are documents the consultant can edit, globally or per individual engagement — the methodological know-how stays in the professional’s hands, not buried in the code. The second: the entire pipeline is idempotent and traceable — every step knows what it has already processed, every output declares how it was produced, and every document carries the methodological note requiring professional validation.

The result: weeks of document work compressed into hours, with a starting quality that the consultant refines rather than builds from scratch.

---

Lekton Dossier is currently in use at compliance consulting firms. Contact us to evaluate it on your own operations.