Interactive Professional Portfolio

Making a professional history searchable, useful and accountable

I started working on the Interactive Professional Portfolio because I was finding it increasingly difficult to represent the range of things I’d done professionally, particularly as I began building more software. I already had a CV, a website and various accounts of individual projects, but none of them really explained how the work fitted together or gave somebody much opportunity to explore it for themselves.

There was also a substantial amount of material I hadn’t looked at for years. Some of it was quite straightforward to recover, such as publishing records, old websites and repositories. Other material was scattered across correspondence, project folders, operational documents and different versions of my own accounts of what I’d been doing.

I wanted to see whether I could bring this material together into something more useful than another extended biography. The idea was to create a body of professional evidence that could support different questions about my experience, including questions I might not have thought to answer when writing a conventional portfolio.

That led me into a much larger piece of work than I had originally anticipated.

Reconstructing the record

One of the difficulties with reconstructing a career is that the evidence varies enormously in what it can tell you.

A publishing record might establish exactly when something appeared and who was responsible for it. An old website can show what a project looked like at a particular point. A repository provides a different kind of history, although having code in a repository doesn’t necessarily tell you whether it ever reached production. My recollections can explain why I made a particular decision or what an experience meant to me, while not always providing reliable dates.

There were also questions about authorship. MoneyScience, for example, involved publishing material from contributors and producing automated information products. I operated the business and its systems, but that doesn’t make every piece of text it published something I personally wrote. Similar distinctions apply to work I’ve edited for other authors and to software developed with AI assistance.

I spent a considerable amount of time comparing material, recovering missing context and deciding which conclusions the evidence could reasonably support. Dates, project relationships, the difference between prototypes and deployed products, and the distinction between contemporary accounts and later recollections all needed attention.

Looking back, this was one of the more interesting parts of the exercise. It meant revisiting work that had gradually disappeared from the accounts I normally gave of my professional life, while also questioning some of the explanations I had become accustomed to using.

When improving the structure made the answers worse

The Portfolio went through several substantial revisions as I experimented with different ways of organising the material and making it available to AI retrieval systems.

One of the more instructive episodes occurred during development of V10. An early version had a much cleaner structure, with stronger distinctions between evidence, interpretation and the various kinds of information the system needed to handle. On paper, it looked like a considerable improvement.

Unfortunately, when we tested it against questions that the previous version could answer, some of the answers were noticeably worse.

The earlier corpus contained detailed observations, dated records and specialist material that had become much less accessible in the new version. The system could explain the broad shape of my professional history quite convincingly, but sometimes struggled to recover the specific evidence behind that explanation.

This was particularly revealing because the new architecture was not obviously defective. It had succeeded in making the information more orderly, which was what it had been designed to do. The difficulty was that we had lost some of the detail that made the information useful.

I decided to return to the previous V9 corpus and use it as the benchmark for the reconstruction. Rather than continuing to refine the new model in isolation, we worked back through the earlier material, identifying what had been retained, what had disappeared and which kinds of questions were being affected.

On 23 September, that became the basis of a substantial reinvention of V10. We went back through the 37 V9 files, reconciled newer evidence and rebuilt the retrieval material with much more attention to whether the original observations could still be recovered.

It changed how I thought about the testing. We needed to know not only whether the files had been processed correctly, but whether the system could still answer the questions that made collecting them worthwhile.

Finding the connections without inventing a story

As the evidence developed, I became interested in what it might reveal about the relationships between different parts of my work.

I had worked in scholarly publishing, quantitative finance, online communities, specialist software platforms, professional associations and a variety of client projects. More recently I had begun designing and implementing software with coding agents. There were some obvious similarities in the kinds of problems involved, although the technologies, organisations and my own responsibilities had changed considerably.

We experimented with ways of identifying recurring capabilities across that history. The process involved testing proposed connections against individual projects, including smaller pieces of work that didn’t necessarily fit the emerging picture.

That was important because a reasonably persuasive account of someone’s career can be constructed by choosing a few suitable examples and interpreting everything else around them. I wanted to be able to challenge those interpretations, including the ones I found attractive.

Some connections survived that process and others became more qualified. There are clear continuities in the way I’ve worked with publishing, information, people and technology, but I don’t want to suggest that the projects were all steps towards something I had already envisaged. The current interpretation is necessarily informed by what has happened since.

The Portfolio now supports several ways of exploring the same underlying history, including individual periods and projects, recurring capabilities, changes in working method and the relationship between earlier experience and current work.

Building and testing the current Portfolio

By the beginning of October 2026, the project had developed into a substantial, versioned collection of professional evidence and supporting analysis.

The current canonical release, V10.4, was promoted on 18 October 2026, following V10.3, which was promoted on 7 October. It contains 160 core professional and career records, together with 26 analytical objects and 19 annual public Twitter/X evidence collections, making 205 objects prepared for retrieval in total.

A considerable amount of the development work has concerned what happens when that material is prepared for use by an AI system. There is a difference between having the right information in a source file, having it available to the retrieval service and obtaining an answer that uses it correctly.

We introduced successive checks to examine those different stages, including questions drawn from earlier corpus versions, tests involving exact dates and project maturity, and cases where the system needed to recognise limits on what could reasonably be inferred.

For the V10.3 release, we assembled a complete verification corpus and checked all 205 objects against the expected inventory. We then ran fourteen acceptance tests using the actual retrieval service. They covered a mixture of professional history, current project states, authorship, privacy and the distinction between different products and systems. All fourteen passed following review of the results.

The release process also separates the material used to answer professional questions from the records used to manage and verify the corpus. Previous releases remain available as historical records, rather than being treated as additional competing sources of current information.

V10.4 is now the canonical corpus, and I have confirmed that it went live through the Niche Clever experience on 18 October 2026. It revised thirteen of the 205 retrieval objects, chiefly to describe the current Niche Clever Interactive product and to treat its earlier generations as history; the other 192 are unchanged from V10.3.

Letting visitors explore the work

The public result of this work is Ask Niche Clever, which gives visitors a way to ask questions about my professional experience rather than relying entirely on the pages I’ve chosen to write.

Someone interested in my publishing background might ask about the years between the Institute of Physics and my more recent software projects. Someone evaluating my technical experience might want to understand what I actually built for Trigger SMARTS, including which versions were used operationally. Another visitor might be interested in how the experience of running MoneyScience relates to the systems I’m developing now.

These questions require different kinds of evidence, and some need information drawn from several periods or projects. That’s one reason I wanted to build the Portfolio in the first place. A conventional website can introduce selected work and explain why it matters, while the conversational interface provides a route into the larger body of material.

I don’t regard the resulting answers as a substitute for examining the evidence, particularly when a claim is consequential or the source record is incomplete. The quality of the interface depends heavily on the work that has gone into preparing and testing the material behind it.

The public Portfolio is also deliberately selective. Some of the material used during research belongs in private archives or internal project records, and the detailed process used to construct and maintain the corpus remains part of Niche Clever’s internal working practice.

A separate piece of software emerged

Developing Ask Niche Clever also involved building a WordPress interface that could handle conversations with an approved knowledge base.

That work led to Niche Clever Interactive, a separately developed, reusable WordPress product. It provides the interface and WordPress-side controls for an existing knowledge base, including configuration, access rules and the practical arrangements needed to make the service usable on a website.

I wanted to keep this distinction clear because the problems are quite different. The plugin provides a way to interact with information that has already been prepared, whereas the Interactive Professional Portfolio is concerned with assembling, evaluating and maintaining that information.

The independently rebuilt Niche Clever Interactive Core and its separate Premium companion have been developed and tested. No WordPress.org approval, directory listing or public download has been recorded for the Core plugin.

My role in the project

I’ve directed the Portfolio from its initial purpose through the evidence reconstruction, successive corpus designs, testing and public implementation.

Much of the research and technical work has been conducted with AI agents, working from documented tasks and shared project context. My role has included investigating sources directly, clarifying historical details, deciding which interpretations to accept, reviewing candidate output and determining when a version was ready to progress.

There has also been substantial practical software development, particularly around the WordPress interface, retrieval integration and the procedures used to assemble and validate releases. Coding agents have made it possible for me to undertake more of that implementation directly, although I distinguish between directing and testing the development and manually authoring every part of the code.

One of the continuing challenges has been maintaining enough documentation and control that work can pass between tools and sessions without losing track of decisions. This has become an important aspect of several other Niche Clever projects, including Harness and Mea Culpa, and the methods have developed alongside one another.

Where the project stands

The Interactive Professional Portfolio is operational, with V10.4 as its current canonical corpus. Ask Niche Clever provides the public conversational experience, while Niche Clever Interactive is the separate reusable WordPress product.

The Portfolio itself remains an active project. My professional work is continuing, and there are still areas of the historical record where the available evidence is incomplete or where a particular interpretation needs further examination.

I’ve found the exercise useful for understanding my own work, but it has also changed how I approach other projects involving large collections of information. The difficulties we’ve encountered with retrieval, inconsistent records, source attribution, privacy and changing interpretations are not peculiar to a professional portfolio. They appear in publishing, organisational knowledge, research collections and operational systems.

Working through them in such detail on my own material has given me a practical way of developing and testing approaches that can be applied elsewhere.