Data Driven CV — The Plan

• Resume-as-data is the established pattern. Keep one canonical structured file set; generate every tailored version from it. JSON Resume is the open standard; a master CV tailored per role is the standard 2026 workflow.
• RenderCV is the strongest render tool (17k+ GitHub stars, Python, actively maintained): CV content in YAML with Markdown bullets, rendered to a typeset PDF via Typst in milliseconds, validated against a schema. Markdown output converts to Word via pandoc.
• Format for sending: clean single column, standard headings, no tables, text boxes or graphics. PDF is the UK default, but .docx is safest for agencies, as every ATS parses it and recruiters reformat contractor CVs onto their own templates (ATS format guide, UK ATS CV guide). The pipeline produces both from the same source.
• The selection step is the differentiator. No off-the-shelf builder chooses the right bullets for a target role well; Claude reads the knowledge base plus a job spec and assembles the flavour. That is the part done here rather than by a tool.

The HTML pages here in Claude Compass are the working view: the Employment History table, one page per engagement, and the research notes. The YAML is the data. Any change agreed on a page is mirrored into the YAML in the same session, and vice versa.

Each achievement bullet carries tags (fabric, power-bi, ai, finance, architecture...) that drive flavour selection, and a relevance rating that captures Matthew's judgement of what matters now:

• Core leads on the main CV: current and differentiating (AI consultancy, Fabric, strategic advisory).
• Supporting included where space allows or a flavour calls for it (stakeholder work, quantified outcomes).
• Background knowledge base evidence only; rarely appears on a CV (legacy delivery mechanics).

• PDF — typeset via RenderCV, the default for direct applications.
• Word (.docx) — for agencies and ATS portals, produced from the same source.
• Online (optional, parked) — an HTML CV at a Disseminate subdomain (e.g. cv.disseminate.co.uk on Cloudflare), phone number omitted, linkable from LinkedIn.

• Implied first person, active voice, everywhere: start with the verb ("Designed and ran enablement sessions..."). Never "I", never passive ("sessions were run" could describe a participant). Exception: scene-setting about the client's situation. Agreed 6 October 2026; full rule in kb\README.md.
• British English; no em dashes or connective hyphens.
• Quantified outcomes wherever possible (£50,000 per month profit, 2 billion observations, 17 divisions, £3.8M gap, under £10 a month).
• Dates and titles follow LinkedIn as displayed unless Matthew confirms otherwise; discrepancies with the old CV are flagged "to confirm", never silently resolved.

1. Capture — old CV extracted, LinkedIn profile fully captured (experience, skills, certifications, recommendations, activity). Done.
2. Knowledge base — populated for all 19 engagements plus profile, skills, certifications and education. Done.
3. Walkthrough — one engagement at a time with Matthew: narrative agreed, bullets rated, engagement page built. In progress (NG Bailey signed off; WYCA drafted).
4. Flavours — define the first flavour configs (main CV, AI-forward, Fabric architect). Not started.
5. Render — install RenderCV, build the pipeline, produce the first PDF and .docx. Not started.
6. Refresh LinkedIn — update skills (currently legacy only), the NG Bailey entry, and the profile statement to match the new CV. Not started.
7. Web CV (later) — online version of the CV, including a client relationship map: a timeline network showing the re-engagements (Spice to Enserve, Metro Rod and Freedom/NG Bailey; Yorkshire Water twice; Inciper three times; Savills renewals). Interview artefact and web centrepiece; deliberately kept OFF the rendered PDF/docx CV for ATS and skim reasons. Agreed 8 October 2026. Not started.