machine viewenglishraw: /cases/second-brain.mdbuild: 8b050972026-08-24T12:37Z
For a consulting team, we built a system that turns day-to-day work into reusable knowledge. Proposals went from 4 hours to 40 minutes, and handoffs from one week to a one-hour meeting. Without depending on someone who is on vacation, has changed teams, or has already left.
6x Faster to prepare a proposal — from 4h to 40 min · 1h To take over a project already in progress — instead of 1 week of onboarding · 18K Records indexed and searchable in the right context · 2700 Meetings transcribed and incorporated into the knowledge base
In a technology consultancy, much of what makes a project move quickly does not live in the code or formal scope. It lives in context: how much a similar project cost, why a technical decision was made, what the client asked for in a meeting, and what changed during delivery. The change was to make the work itself feed a shared company memory.
Shared knowledge base: Projects, proposals, documents, meetings and decisions become part of the same operational memory, with enough context to be found and reused later.
Automatic work capture: Meetings, conversations and existing records feed the knowledge base automatically. Knowledge grows as a by-product of the work, not as an additional documentation task.
Living project documentation: Context and decisions are recorded throughout delivery. Someone joining later can understand not only what was decided, but why.
AI inside the workflow: Assistants use accumulated knowledge to retrieve similar projects, summarize meetings, turn conversations into next steps and keep records up to date.
Boundary between company knowledge and private information: Not everything a person knows should become shared knowledge. The system defines what can be incorporated into the knowledge base and what remains restricted.
Controlled access to knowledge: AI does not receive unrestricted access to the operation. People and assistants can only query information permitted for that specific context, with dedicated rules for sensitive data.
Related offering: [AI in the Operation](https://bleu.builders/offerings/ai-transformation/)
Contact: [Start a conversation](https://bleu.builders/contact/)
Consulting · AI IMPLEMENTATION · LIVE
Consulting
For a consulting team, we built a system that turns day-to-day work into reusable knowledge. Proposals went from 4 hours to 40 minutes, and handoffs from one week to a one-hour meeting. Without depending on someone who is on vacation, has changed teams, or has already left.
6x
Faster to prepare a proposal — from 4h to 40 min
1h
To take over a project already in progress — instead of 1 week of onboarding
18K
Records indexed and searchable in the right context
2,700
Meetings transcribed and incorporated into the knowledge base
Overview
In a technology consultancy, much of what makes a project move quickly does not live in the code or formal scope. It lives in context: how much a similar project cost, why a technical decision was made, what the client asked for in a meeting, and what changed during delivery. The change was to make the work itself feed a shared company memory.
Our role
Before implementing AI, we mapped how the operation actually worked: where information originated, how it moved, who needed it later, and what was simply a record with no future value. From there, we designed the system around three principles:
- Capture knowledge without creating a new documentation burden for the team.
- Connect information created in different places.
- Bring that knowledge back at the exact moment it becomes useful again.
That also meant deciding what should become shared company knowledge, what should remain private, and which information AI could access.
What changed
The system began capturing context as the work itself happened: recorded meetings, synchronized conversations, documents and records already part of the team’s routine automatically feed a shared knowledge base. Each completed project therefore reduces the effort required to start the next one.
- The knowledge base brings together thousands of records and meetings without requiring someone to stop later and ‘document the project’.
- Information is no longer simply archived; it resurfaces when there is a decision to make.
- When preparing a proposal, the team retrieves pricing, scope and context from similar projects; work that used to take around 4 hours now takes approximately 40 minutes.
- When taking over a project in progress, decisions, history and context are already organized; a week of onboarding has been replaced by reviewing the material and a roughly one-hour meeting.
What we've shipped
The first visible change appeared in the proposal workflow. The remaining layers were introduced progressively, without interrupting projects already underway.
Shared knowledge base
Projects, proposals, documents, meetings and decisions become part of the same operational memory, with enough context to be found and reused later.
Automatic work capture
Meetings, conversations and existing records feed the knowledge base automatically. Knowledge grows as a by-product of the work, not as an additional documentation task.
Living project documentation
Context and decisions are recorded throughout delivery. Someone joining later can understand not only what was decided, but why.
AI inside the workflow
Assistants use accumulated knowledge to retrieve similar projects, summarize meetings, turn conversations into next steps and keep records up to date.
Boundary between company knowledge and private information
Not everything a person knows should become shared knowledge. The system defines what can be incorporated into the knowledge base and what remains restricted.
Controlled access to knowledge
AI does not receive unrestricted access to the operation. People and assistants can only query information permitted for that specific context, with dedicated rules for sensitive data.
Areas we run
PROPOSALS IN MINUTES, NOT HOURS
In a consulting firm, time spent preparing a proposal is time that has not yet turned into delivered work.
Proposals stopped starting from scratch
Before, preparing a proposal meant finding an old reference, remembering who had worked on something similar, and manually reconstructing pricing, scope and context. Now, the system finds similar projects that have already been delivered and brings forward what is relevant to the new opportunity. A proposal becomes an adaptation of existing knowledge, not a reconstruction.
- Pricing, scope and context from similar projects ready to reuse
- History retrieved from the knowledge base, not from someone's memory
- Preparation time reduced from around 4 hours to 40 minutes
HANDOFFS THAT DO NOT DEPEND ON WHO LEAVES
Vacations, team changes or someone leaving the company should not mean losing project context.
A one-hour handoff, not a one-week onboarding
Before, an important part of a project's knowledge existed only with the people working on it. When someone needed to take over delivery, the reconstruction began: meetings, messages, documents, decisions and questions for the person leaving. Now, that context is produced throughout the project. The person taking over finds history, decisions and documentation already organized and uses the handoff meeting to clarify exceptions — not to rebuild everything.
- Documentation ready before the handoff starts
- Decisions recorded together with the reasoning behind them
- A roughly one-hour meeting replaces a week of onboarding
- The project keeps moving when someone goes on vacation, changes teams or leaves the company
AI THAT CAN QUERY THE COMPANY WITHOUT SEEING THE ENTIRE COMPANY
The more useful an internal assistant becomes, the more important it is to define what it must not know.
AI access is constrained by design
Centralizing knowledge created a new responsibility: making sure a simple query could not surface information that the person asking should not be able to access. Permissions were therefore not added as an afterthought. The assistant can query only information and paths authorized for that context. Sensitive data remains subject to the same access boundaries that apply to people. AI implementation stopped being only a question of model capability and also became a question of governance, context and boundaries.
- Access defined according to role and context
- Sensitive information remains outside unauthorized queries
- The assistant works with real operational knowledge without receiving unrestricted access to the entire knowledge base
RELATED OFFERING
See how Bleu structures this kind of work
The case shows what happened in one partnership. The offering explains the entry point, scope, and first delivery.
Bring us the problem.
Tell us what is going on. After you send it, the calendar opens so you can choose 15 minutes.
15 MINUTES · A CLEAR NEXT STEP.