Ask questions about the work published on this site.

Ask a question about the published work here — the evidence write-ups, the CV, the writing. The answer is built only from this site's pages, and every claim links to the page it came from. The model runs in this browser, so nothing is uploaded.

First run downloads a model. The language model is fetched from Hugging Face the first time you run this — about 900 MB, so use Wi-Fi — and cached in your browser afterwards. The tool code is loaded from the jsDelivr CDN, and your questions are not sent anywhere.

Ask a question

Loading the corpus…

Checking whether this browser can run the model…

Example answer · load the model to ask your own questionHe technically led a multidisciplinary team on a national transport dataset, including an event-driven ingestion redesign and a guarded workflow for permanently removing records. [1]

[1]Leading safer change in a national transport dataset — evidence

How it works. Your question is matched against passages from this site's pages using BM25, a keyword-ranking method that needs no model and no network. Only the best-matching passages go into the prompt, and if nothing matches well enough, the tool says so instead of guessing.

Technical note: the answer model is Qwen2.5-1.5B-Instruct, run by WebLLM through WebGPU, the browser feature that uses the device's graphics processor. Retrieval and scoring run locally in JavaScript; the corpus is generated from the site's own pages at build time.