private twin
Language FR EN

The product

What exists today

Private Twin is an advanced prototype in validation. This page shows what it produces, what it looks like, and what is left to prove.

How it works

A reconstruction, not data entry

The agent observes the sources you authorise, attaches activity to the right projects, and rebuilds the current state of each. Nothing is asked of the user during the day: what is rebuilt is presented afterwards, for approval or correction.

Observe

The authorised traces of work, and only those: mail, calendar, files, project tool.

Attach

Link activity, meetings, communications and files to the right projects, people and clients.

Rebuild

Recompose the current state of the record: decisions, open actions, risks, evidence, effort.

Propose

Present the result as a proposal. Approving, correcting and refusing stay with the user.

Outputs

Three outputs, one engine

Time

A defensible timesheet. Effort recovered, attributed to the right project, with the evidence that backs it.

Time is an output, not the promise. What is aimed at is a state of the project that is usable, sourced and justifiable.

State

A daily readout per project. What changed, what is moving, what is blocked, what is about to slip.

Actions

The actions to follow. Tasks and promises detected in real work, proposed for your approval and kept with their context.

All three come from the same execution record. You are not installing three products.

Screenshots

What it looks like

Three screenshots from the prototype running today. They do not show a finished product: they show that the reconstruction exists and produces readable output.

Private Twin screenshot: the reconstructed timeline of a working day, time attributed to projects, meetings detected, and unattributed periods still visible
Prototype · reconstructed timelineA reconstructed day: time attributed to projects, meetings detected, and the periods the system could not attribute, left visible rather than filled in.
Private Twin screenshot: the thread of a day, events grouped by source and reconstructed links between those taking part in the same work sequences
Prototype · a day of interactionsThe thread of the day: events grouped by source, and the reconstructed links between those taking part in the same work sequences.
Private Twin screenshot: a proposed narrative memory, marked as a decision and reviewed, attached to a client and a stakeholder, with its sources and the correct or archive actions
Prototype · sourced memoryA memory proposed from the traces, attached to the right client and the right person, with its sources. Correcting or archiving stays with the user.

Difference

What these tools are not built to do

Each of these tools does one thing well. None of them is built to keep a project’s state current from the work actually done.

AlternativeWhat it does wellPrivate Twin
Assistants and copilotsDialogue, reasoning, generation, connectors.Operational context already reconciled, attributed, sourced and durable.
Domain RAGDocument search, ontology, answers and deliverables.Real execution beyond documents: timeline, interruptions, effort, open actions.
Time trackersAutomatic capture and attribution of time.A record linking effort, decisions, actions, risks and evidence.
Tasks, PM, CRMExplicit recording of tasks and statuses.Detection of the gap between what is declared and what actually happened.
Process miningAnalysis of structured transactions.Reconstruction of knowledge work from heterogeneous, uncertain traces.

These players can technically converge. The difference is only defensible if Private Twin’s outputs are more reliable thanks to real activity and human validation.

Getting started

What it actually asks of you

  1. 01

    Authorise the sources

    You choose what the agent observes: mail, calendar, files, project tool. Nothing else. Each source is switched on and off separately.

  2. 02

    Let it rebuild

    The first days go into attaching activity to the right projects. What is rebuilt is proposed to you, never applied behind your back.

  3. 03

    Correct it

    Your corrections are the tuning. They then make up most of the data entry asked of you, and they shrink as the attribution gets it right.

How much that learning actually costs is part of what is being validated. It is one of the reasons I am looking for pilot teams rather than users.

Status

Built, and being validated

Built

  • A complete prototype that collects, links and interprets different traces of work.
  • An architecture designed to preserve context, sources and human approvals.
  • A harness that limits dependence on language models and their providers.
  • A working Sim Lab for monitoring, replay and version comparison.

Being validated

  • Evaluating results across more days and more real situations.
  • Improving features before switching them on for users.
  • Identifying which customer profiles feel the problem most sharply.
  • Validating usage, adoption and the value actually created.

Test the reconstruction on your own projects

I am looking for professional services teams of 5 to 50 people willing to run the reconstruction on their own projects, and tell me where it gets things wrong.