Case studies, on a separate site
The systems, drawn at the level an interviewer asks about.
A separate case-study site documents the AI systems I contributed to at BP-ITCS: twenty architecture diagrams, an interactive walkthrough and a written record of what each claim rests on. It is published on its own terms, with its own review of what could be shown.
01The case studies
The case studies
Every one of these is on the separate case-study site. The class beside each title is the contribution level that site publishes for it, and two of the seven are deliberately not claimed as mine.
02Ways in
Where to start
The first is the whole thing. The other three are the questions it gets asked most, each opening at the section that answers it.
- Start hereOpen the interactive architectureOne page, every diagram inlined, with a control that dims everything I did not implement so authorship is visible across the whole system at once.Browse
- Technical interviewRun the interview walkthroughAn eleven-step guided sequence through the same page, in the order the work is best explained out loud: the system, then one document through it, then the checks, then what is mine.About 20 minutes
- Engineering depthRead the verification argumentThe part worth the most time: how a stored representation is checked back against the bytes the publisher served, and why a checker that shares its subject's assumptions cannot falsify anything.About 10 minutes
- Scope and honestySee the contribution mapEvery repository classified by what I actually did in it, including the ones where the answer is nothing. Commit shares are given as evidence of weight, not as proof of authorship.About 3 minutes
03Scope
What this covers
Worth reading before the diagrams rather than after. The case-study site makes a narrow claim carefully, and the narrowness is the point.
What it is
- Architecture case studies of systems I contributed to as an engineer.
- A stated contribution level for every component, separating what I implemented from what I extended and what was already there.
- Claims traced to files, with the method written down rather than asserted.
What it is not
- Official documentation of any employer platform.
- A claim to have built a corporate platform single-handed.
- Deployment detail, credentials, hosts, addresses or customer data, all of which were removed rather than obscured.
04The other treatment
Why there are two versions
This site and the case-study site publish the same employer work at different levels, and it is worth knowing which one you are reading.
On this site the work appears as Reliable knowledge systems, which is synthetic throughout. An illustrative model of a class of problem, not a description of any deployed system. Every quantity is synthetic and chosen for legibility. No architecture, technology, identifier, or scale from any employer system appears anywhere in it. It carries the class of problem: which invariants a system of that shape has to be able to check about itself, and which failures those checks exist to make visible.
The case-study site carries the particulars, published after its own review of what could be shown and its own separation of public material from private evidence. If you want the argument, read the synthetic model. If you want the system, open the case studies.