I’ve never known quite how to respond when people ask me what I do. I can usually give them an answer appropriate to the circumstances. Publishing, consulting, building websites, working with people and information. Each describes something I’ve done, and the work often involves several of them at once. The difficulty is explaining how they fit together well enough for someone to recognise where I might be useful.
That’s one of the reasons for rebuilding Niche Clever. I want the website to give people a clearer sense of what I can do and how I might be useful to them. I also want it to demonstrate what has become possible through my work with AI, and provide a place to explore the questions that work raises.
What AI has changed for me
AI has made an enormous difference to my capacity to turn an idea into something that works. I can build tools, investigate problems and develop software at a scale and pace that would previously have been beyond me. Something I might have wanted to do for someone, but couldn’t find the time or capacity for, can now become possible.
That capacity draws on experience accumulated across publishing, editorial work, organisational operations and technical projects. Knowing what a system needs to do involves understanding the people using it, the information passing through it, and the decisions it has to support. AI has changed how much of that understanding I can put into practice myself.
The website is intended to demonstrate this. Its construction, the tools it brings together and the projects it presents are examples of what I can now make. I want visitors to have enough substance to judge that for themselves. A working example can explain something that is difficult to convey in a description of services.
Making my professional history queryable
The Interactive Professional Portfolio has become central to that effort. The initial question was straightforward enough: could I make my professional history something people could ask questions about?
The material for answering those questions was scattered across CVs, websites, project files, publishing records, correspondence and surviving fragments of earlier work. Bringing it together involved recovering information, comparing accounts and deciding what each source could establish.
Going back through those records was fascinating. There was a continuing exchange between the material and my recollections. An old document could bring back a project or clarify its sequence. A question about that document could reveal something I remembered which the surviving record didn’t explain. That clarification would change the account, and sometimes lead to another search. The process also gave me a different perspective on how the work connected across periods.
An important question was how much detail to preserve. A career summary might establish my role at a particular time. A richer record could support questions about the work passing through that role, the people involved and the interests I was pursuing alongside it.
The corpus includes tweets from periods when I was active on Twitter. It also contains publication records from my time working on Quantitative Finance. Those can help reconstruct which articles were in the surrounding editorial workflow during a particular period. The precision depends on the evidence: a publication date and a known responsibility can support a useful account of a month without establishing what was on my desk on a particular day.
That level of resolution makes different kinds of questions possible. It also means paying attention to what the material actually says, who produced it and what conclusions it can support.
A period of research and development
A broken pelvis put me in hospital for two weeks while this work was going on. A substantial amount of the corpus refinement happened during that period. It became an extensive piece of research and development into how to construct a body of knowledge that an AI could use effectively.
Building the professional record required tools to support its construction. There was material to recover and organise, versions to assemble and compare, corrections to carry forward, and decisions to preserve so that work could continue across models and conversations. Each part needed enough context for the next piece of work to make sense.
It also required a great deal of testing. I needed to understand what level of detail the corpus should contain, how it should be arranged, and whether the resulting answers could actually use it. Historical research, my own recollections, editorial judgement, information modelling and tests of retrieval and interpretation all contributed. The approaches developed together as the work progressed.
At the time of writing, the current version is 10.2, which is live, and contains around 160 documents. Its governance is complex because it has to handle different kinds of material and different kinds of claim. The principles behind it are fairly easy to state: keep track of sources, preserve relevant context and chronology, make uncertainty visible, distinguish evidence from interpretation, respect privacy and authorship, and retain the ability to review and revise.
Those principles influence what goes into the corpus and how an answer should use it. A text produced by a publishing system I operated can demonstrate something about the system and my work on it. Establishing that I personally wrote the text is a separate question. A project record may show that software was built, while adoption by a client requires further evidence.
The testing involved asking ordinary questions and looking closely at the answers. Could the system recover an actual quotation? Could it explain what was known about a particular period? Could it discuss connections between projects without turning a later interpretation into a claim about what I had intended at the time?
I also needed to examine whether the evidence had survived the process of organising it. A corpus can retain a description of a collection while losing the detail that made it useful. The same care applies to the answers: a fluent account can overstate what its sources establish.
This gave the work a continuing cycle of construction, examination and revision. The outcome was a much richer professional record, together with tools and a developing method for maintaining and questioning it.
Bringing the portfolio into the website
The portfolio now supplies much of the material from which the site can present my work. The case studies give an authored account of particular projects. The corpus carries a broader record, with chronology, supporting evidence and relationships between projects. Ask Niche Clever provides a question interface over that material.
Building the corpus, building the interface and writing the website each involve decisions about what someone needs to understand. They bring different approaches to the same material, allowing a visitor to move between an overview and a more detailed examination.
I want those routes to accommodate different kinds of curiosity. Someone may want a quick sense of what I do. Someone considering working with me may want to investigate a relevant project, understand my contribution or ask about comparable experience. Someone else may be interested in how the portfolio itself was constructed.
The site can therefore demonstrate both the work I have done and an example of what I can now build. Refining the answers and maintaining the material behind them remain continuing work.
Making room for work and inquiry
The blog gives the site another continuing purpose. I want to explore the philosophy, ethics and practical use of AI, and build a readership around that inquiry. Its effect on my own life has been profound. My concerns about its wider consequences remain substantial, including environmental costs, creative labour, inequality, psychological effects and the institutions responsible for overseeing it.
The next piece is a short manifesto for engaging with AI. A philosophical essay examines the reasoning and tensions behind it. After that, a series of deeper discussions will take individual principles through examples, objections and research, with scope to revise the manifesto as my understanding develops.
I want the site to be the continuing home for that writing, with ways for readers to follow it and encounter it elsewhere. Paid projects can also contribute to this ongoing research, bringing real problems, constraints and people into the work.
What I hope the relaunch makes clearer is how I might be useful to someone else. That might begin with a publishing workflow, information scattered across several systems, a recurring administrative task or an idea that needs a working prototype. The site brings together examples, experience and ways to explore them, so that someone encountering the work can begin to recognise what we could do together.
In this series
Next in this series: A short manifesto for engaging with AI.
This piece was developed in conversation with AI. The ideas, experience and decisions are mine; AI helped draft and revise the wording, which I reviewed and accepted.