Technology
How Gilbert learns the vocabulary of your industry
Corporate law, EBITDA, sprint backlog: how our models natively recognise the jargon of more than ten professional verticals.
"Clause résolutoire", "adjusted EBITDA", "sprint backlog", "anamnesis". Every industry has its vocabulary, and it is precisely that vocabulary which carries the meaning of your professional conversations. Here is how Gilbert recognises it.
The problem with general-purpose models
A general-purpose speech recognition model is trained on consumer content: podcasts, radio shows, YouTube videos. It knows everyday words, but not professional jargon.
The result: when a lawyer says "nantissement de parts sociales", a general-purpose model may transcribe "en en tissements de par social". That is unusable.
Our approach: sector language models
For each professional vertical we assembled what we call a sector language model, a specialised model that knows:
- the vocabulary of the sector, thousands of technical terms and their variants
- the frequent collocations: in French, "clause de" is far more often followed by "non-concurrence" than by "non-compétence"
- the patterns of conversation: in an investment committee, certain sequences of terms are predictable
- the named entities: laws, standards, institutions, financial indices
More than ten verticals covered
Gilbert covers natively:
- Legal: corporate law, employment law, litigation, compliance
- Finance and banking: financial analysis, asset management, audit, insurance
- Consulting and strategy: management consulting, digital transformation
- Executive management: governance, strategy, mergers and acquisitions
- Sales: B2B selling, CRM, pipeline management
- Real estate: transactions, property management, development
- HR: recruitment, training, employee relations
- Technology and product: engineering, product management, DevOps
- Healthcare: medical vocabulary, pharmacology
- Insurance and compliance: underwriting, claims, regulation
How it works in practice
When you start Gilbert, our system detects the sector context in the first seconds of the conversation. It then activates the right sector model, which significantly raises recognition accuracy on technical terms.
There is nothing to configure and nothing to set up. It is automatic.
What if my sector is not covered?
Our base model still performs well outside the covered verticals. And if you are an Enterprise customer, we can train a sector model for your specific field, in a matter of weeks.
The vocabulary of your profession is not an obstacle. It is our speciality.
