A Shared Knowledge System for Specialized AI Workflows

The problem

I use specialized AI assistants for different kinds of work, including coding, career planning, and personal training. Specialization made each workflow more useful, but it also split context across separate conversations. When one assistant had project details another needed, I had to repeat or copy that information between sessions. That added friction and duplicated context instead of letting each assistant build on work already done.

I wanted a shared knowledge system that could preserve useful decisions across sessions without letting every assistant freely rewrite the common source of truth.

The design

I chose a shared Obsidian knowledge base as the common reference. Specialist workflows can read shared knowledge for context, but each contributes only to its own branch. A designated operator reviews the branch notes and daily or weekly summaries, organizes durable information, and maintains the shared layer.

If the operator encounters conflicting facts, it asks me to decide which account is accurate before changing the shared record. The system is designed to preserve continuity, not to claim perfect or automatic recall: useful information still has to be captured, reviewed, and maintained.

How the system works across roles

The same knowledge flow supports distinct personal workflows. A fitness specialist can use relevant retained context for workout planning. A career specialist can work from organized career and project information. A coding specialist can record verified project details in its branch for later use.

One example is the project-to-resume handoff: the coding workflow documents project evidence in its own area; the operator reviews and organizes verified information into the shared, approved record; then the career workflow can use that evidence to prepare resume material. I no longer have to copy and paste the same project context between specialist conversations for each handoff.

My contribution

I defined the system’s purpose and specialist responsibilities, chose the shared knowledge approach, set the read/write boundaries, and established the rule that I resolve conflicts before contested information becomes part of the shared record. I also directed workflow and maintenance decisions and reviewed whether the system behaved as intended. AI assistants substantially helped implement, document, and verify the system; I am not claiming to have personally written every line of code.

Impact and limits

In my own use, the shared workflow reduces repeated explanations and manual context transfers, and helps me move from project work to career materials without reconstructing the same background each time. Avoiding repeated context transfers is also part of the system’s cost-aware design. These are qualitative benefits, not measured outcomes: I have not tracked exact time saved, token reduction, or a before-and-after baseline.

The system is a private personal workflow, not a product deployed to external customers. Its value depends on keeping source notes current and having the operator reconcile conflicts; it does not guarantee that every conversation or detail will be remembered.