Choices
An AI product manager I co-founded to help founders decide what to build, turning customer research into prioritised opportunities, product specifications and development tasks.
Visit Choices (opens in a new tab)How Choices evolved
Choices began as a way for founders and builders to find collaborators. It has since grown into a product-planning workspace. A founder can start with an idea, add interviews or market research, and work towards a specification that explains what to build and the evidence behind it.
What I built
As co-founder and engineer, I built and deployed the public beta using Next.js, React and Tailwind CSS. My work covered LLM API integration, connecting generated research to source evidence, and role-based access for shared projects. I also implemented task exports to help users move from planning into development.
Keeping the research attached
The pipeline moves from evidence to clusters, opportunities and specifications. It groups related evidence using embedding similarity and retains references to the original items. A scoring function weighs demand, urgency, revenue potential and feasibility to help prioritise opportunities. Supporting quotes are passed into specification generation, while a separate mode asks the model to label assumptions when real evidence is missing. The scores are prioritisation heuristics, not forecasts of commercial success.
From specification to delivery
Specifications include features, acceptance criteria, risks and release conditions. The Linear integration creates issues with their supporting context and records export IDs to prevent duplicates. For coding assistants, the task board produces a Markdown pack with task descriptions and acceptance checklists that users can copy into tools such as Cursor or Claude Code.
Tools
Next.js · React · TypeScript · Tailwind CSS · Supabase · LLM APIs