2026 · 3 months · Personal project
AI Design Technologist, Product Designer, Full-Stack Developer
Local-first job search copilot that tailors a CV and motivation letter to every application, scores the match, preps the interview, and tracks the whole pipeline. No subscription and no cloud: it runs on your machine, and the AI you already have pays for the computation. AI does the drafting; the human always owns the final word.
My own job search. Every application needed a CV and a letter rewritten for the posting, a judgement about whether it was worth applying at all, interview preparation, and a record of where everything stood. The tools that do this rent you a subscription and hold your data on their servers.
Local-first, by decision: the profile, the documents and the pipeline live on the user's machine, and the AI they already pay for does the computation. No account, no server, no subscription. That single choice sets the privacy model, the cost model, and the hardest adoption problem the product has, all at once.
A FastAPI and React desktop app built around one profile as the source of truth. Relevance-driven generation: each CV and letter pulls from the complete profile only what the specific posting needs. Generation is constrained rather than trusted, the model fills a validated design spec instead of writing the layout, so every CV compiles and a contrast guard keeps it legible whatever the model proposes. Preserve-by-default editing with regeneration previews and lockable sections, so nothing is overwritten without showing the change first.
Solo. Product strategy, user research, design system, architecture, backend, frontend, Typst document templates, and the Playwright E2E suite.
Shipped and in daily use: it runs my own job search, which is also how every feature got tested. Around 400 commits, 30+ written feature plans and a dated product-decision log over three months.
Status: Shipped, not sold. Validation gate not met.
You do not rent Flarekeep; you own it, and the AI you already pay for does the computing. The data, the documents and the pipeline already live with you.
An honest caveat: at times I felt I was building a solution for a pre-AI world. Soon everyone will draft their documents with AI and recruiters will screen with AI, and artifacts like the motivation letter will lose their value or mutate into something else. Part of what Flarekeep automates today has an expiry date. I still believe the pipeline, the data ownership and the orchestration survive that shift, but knowing which layers are temporary changed how much I invested in each one.