Product Pilot
AI copilot that generates PRDs, architecture docs, and success metrics from conversational input.
The Problem
Product documentation is the bottleneck between ideas and implementation. Writing a PRD takes hours of structured thinking — user stories, technical constraints, success metrics, edge cases. Most engineers skip it or produce thin specs that create ambiguity downstream. AI can generate text, but generic outputs lack the structured reasoning that makes a PRD actionable.
What I’m Building
Product Pilot is a conversational interface that transforms product thinking into structured documentation. You describe what you want to build in natural language, and it generates interconnected document sections: PRD with user stories and acceptance criteria, technical architecture with component diagrams, success metrics with measurement plans, and risk assessment with mitigation strategies.
The key differentiator is document coherence. Each section references and cross-validates the others. A user story in the PRD maps to specific components in the architecture doc. Success metrics tie back to user stories. Risk items reference architectural decisions. This interconnection is what separates useful documentation from generated filler.
Document Generation Pipeline
The generation pipeline works in stages rather than producing everything at once. Stage 1 extracts core requirements from the conversation — what problem, who it’s for, what constraints exist. Stage 2 generates user stories with acceptance criteria. Stage 3 builds technical architecture informed by the user stories. Stage 4 derives metrics from the acceptance criteria. Each stage validates against previous stages, catching contradictions before they propagate.
Iterative Refinement
Generated documents aren’t final. Product Pilot supports conversational refinement — “make the auth section more detailed”, “add an offline mode requirement”, “what if we used WebSockets instead of polling”. Each refinement propagates through dependent sections. Adding an offline requirement updates the architecture (local storage, sync strategy), metrics (offline session tracking), and risk assessment (conflict resolution complexity).
Technical Approach
The system uses Claude’s extended thinking for multi-section reasoning. A single generation pass considers all document sections simultaneously, maintaining internal consistency. Template structures enforce format standards (numbered user stories, consistent heading hierarchy, metric definition format) while allowing content flexibility. Output supports Markdown, Notion export, and structured JSON for programmatic consumption.