Portfolio case study
Signal to Structure
The product problem is that users often ask AI agents for complex work with too little structure. IntentForge solves this by turning rough intent into a prompt framework selected for the task category.
Architecture
Browser input flows through Next.js validation, privacy-preserving rate limiting, task classification, framework selection, missing-context analysis, a replaceable AI provider, schema validation, quality scoring, and safe markdown rendering.
Privacy Decisions
Anonymous prompt content stays in IndexedDB. Signed-in cloud history requires explicit saving and is isolated by Supabase RLS. Usage events store operational metadata only.
Tradeoffs
The MVP favors deterministic mock generation for development, provider abstraction for future model swaps, and simple native controls over a heavy custom editor.
Roadmap
Future releases can add custom frameworks, version diffs, bring-your-own-key generation, team workspaces, prompt templates, and API access.