Context engineering
Turning sixty-five gigabytes of drive dumps into context an assistant can use
AI assistants are only as good as the context you can hand them, and the context was spread across drives, chats and mailboxes. I turned it into a structured base where every past engagement is a self-contained dossier.
- Unpacked, classified and placed in a single pass
- ~1,600 filesUnpacked, classified and placed in a single pass
- Stakeholders · solution · money · technology · what was delivered and when
- Fixed dossier formatStakeholders · solution · money · technology · what was delivered and when
- Every bulk move logged in a manifest before it happened
- Fully reversibleEvery bulk move logged in a manifest before it happened
The client
A consulting and delivery practice with several years of proposals, contracts, estimates, transcripts and reference material accumulated across external drives, shared folders, chat exports and two mailboxes.
The engagement
A single structured pass over the whole estate — split, classify, rewrite as dossiers, and leave behind an intake playbook so it stays true.
The problem
Years of proposals, contracts, estimates, transcripts and reference material lived across external drives, shared folders, chat exports and two mailboxes. That is survivable for a person with a good memory and fatal for an AI-assisted way of working: an assistant is only as useful as the context you can put in front of it, and none of this was retrievable, comparable or safe to hand to anything.
What I did
I split the estate in two — notes in one root, binaries in another — and structured the notes so that finding something does not depend on remembering where it was filed. Every past engagement was then rewritten to a fixed dossier format: who the stakeholders are on both sides and who decides, what the solution actually was and what was explicitly excluded, the money by phase and role, the technology decisions and why they were taken, and a dated log of every artefact sent to the client. That last section is the one everyone skips and the one that makes a dossier usable a year later. Bulk moves ran through a reversible manifest rather than hopeful drag-and-drop, and the sweep surfaced live credentials sitting in forgotten folders — flagged for rotation as a finding in its own right.
What was built
A structured knowledge base with notes separated from binaries, a consistent project and area hierarchy with hub pages, every past engagement rewritten to a fixed dossier format — stakeholders and who decides, solution and explicit exclusions, money by phase and role, technology decisions and why, and a dated log of every artefact sent to the client — plus a reversible move manifest and a written intake playbook.
On the table at the end
- Structured knowledge base with notes and binaries separated
- Engagement dossiers to a fixed section set
- Reversible move manifest
- Intake playbook with weekly triage
- Credential findings flagged for rotation
What it changed
Made years of commercial history retrievable and comparable rather than merely stored, so a past engagement can be answered from in minutes — and the same sweep surfaced live credentials sitting in folders nobody had opened in months.
How it ran
- 01
Split notes from binaries
One root for knowledge, one for files, with pointers from each note to its source material — so the knowledge base stays small enough to read and to search.
- 02
One structure, applied everywhere
Projects, areas, meetings, decisions, risks, playbooks and templates, with hub pages per project rather than a flat pile of notes.
- 03
Rewrite engagements as dossiers
A fixed section set per engagement, deep-read from the estimates and contracts themselves rather than summarised from filenames.
- 04
Move reversibly
A manifest logged before every bulk operation, so a classification decision could be undone rather than argued about.
- 05
A repeatable intake path
A written playbook for new material — arrive, summarise into a source note, tag, link, route — with a weekly triage.
Other work
All case studies →- Professional services
A project office that lives in a git repository
Project knowledge lived in people's heads and chat threads, and AI tools were producing confident documents with no provenance. I made the repository the single source of truth and gave every project management role a written protocol.
- Software services
Nineteen estimates, one register, five minutes to an answer
Four years of pricing lived in spreadsheets nobody could open. I rebuilt the commercial history into one register — what was estimated, what it sold for, what happened next — and found that half the formulas were already dead.
- Enterprise software
Data and people existed; the process and the cockpit did not
A delivery assurance function owned everything and controlled nothing. I diagnosed nineteen gaps, designed the target operating model, and put stop conditions in the roadmap so it could not become another initiative nobody uses.
Something similar on your plate?
Thirty minutes, no deck. I will tell you whether it is worth doing at all.