Conversational intelligence · research
A conversation is more than its words.
Lilith is a Cosen Lab research project on conversational intelligence that adapts the form, conciseness, and depth of its answers to the user’s context and moment, even when several people are in the conversation.
Conversation
IllustrativeThree points, briefly:
- Deadline confirmed for the end of the month.
- The volume clause is still open.
- The client expects an answer today.
Want the full details after the meeting?
Not just what is being asked, but how and when the answer is needed.
Beyond the next word.
Toward the right answer, at the right moment.
Working with an AI shouldn’t mean getting the same kind of text every time. Lilith treats the conversation as a shared space, where pace, roles, and circumstances matter too.
The moment
Someone with two minutes and someone with an afternoon don’t need the same answer.
The roles
A room holds different people, with different questions and responsibilities.
Continuity
Long sessions change direction: the underlying goal stays, the form adapts.
Same question.
Different answers.
The content stays accurate and consistent; what changes is conciseness, depth, and tone. Pick a context and see how the shape of the answer changes.
The same question
In one line: supplier B is more reliable, A costs less.
I’d go with B if continuity matters more than price.
How the answer changes
Tone
Direct
Illustrative example
When many people
are talking.
Operational meetings, negotiations, communication in sensitive situations: Lilith explores how an AI can support dialogue among multiple people and agents without talking over anyone.
One room · illustrative diagram
Orderly turn-taking
Everyone has room to speak; the AI steps in when needed, never over others.
The goal stays central
A change of pace or topic doesn’t lose sight of what the discussion is ultimately for.
Role-based summaries
The same outcome, framed so it is useful to whoever has to act on it.
Adapt to people.
Never manipulate them.
An AI that adjusts how it communicates must do so transparently, respectfully, and in line with European rules.
Transparency
People using Lilith know they are talking to an AI, and why the answer takes the shape it does.
User control
Style preferences and level of detail can always be changed by the user.
No manipulation
Adaptation exists to be better understood, not to steer decisions or behavior.
Compliance
Designed in line with the EU AI Act and the GDPR, with particular care for workplace settings.
Lilith is
in the research phase.
This page describes a line of work, not an available product or measured performance. Results and specifications will be shared once they are mature.
- 01
Concept
Defined
- 02
Lab prototypes
In progress
- 03
Testing
Next phase
- 04
Pilot partnerships
By invitation
Confidential details
To protect the work in progress, models and techniques are not public. We discuss them with research partners and interested companies under a non-disclosure agreement.
Being understood is a form of respect.
Lilith · conversational intelligence. Emerging research by Cosen Lab.