Context
Voters struggle to objectively compare candidates' platforms in municipal elections — and most LLM-based assistants leak political bias.
Solution
Nuxt 4 + Nitro full-stack app comparing 9 mayoral candidates across 184 official proposals in 10 policy categories. Personalized affinity quiz (rule-based, deterministic). AI chat grounded by injecting the relevant proposals straight into the system prompt — RAG by structured DB injection, not embeddings — with a hard neutrality constraint and a Kimi k2p5 backend (Anthropic-compatible API) chosen specifically to step outside the dominant-LLM training-data bias.
