Hackbot

[ReehalS]

Add a LLM based helper to hub.hackdavis.io for hackers so they can find info more easily

[ReehalS]

HackBot is based on the following:

• Knowledge base: Two MongoDB collections: 1. hackbot_knowledge (human-readable source docs with type, title, content, url; managed via admin CRUD at /admin/hackbot); 2. hackbot_docs (vector-embedded versions for semantic search). Events are seeded from the live events collection; all other docs (judging, submission, FAQ, general, track) are managed in hackbot_knowledge and embedded on save.
• Context retrieval: getHackbotContext.ts does adaptive MongoDB Atlas vector search (5–30 docs depending on query complexity) using hackbot_vector_index, with live event hydration from the events collection to keep schedules current.
• LLM: OpenAI provider only with gpt-4o-mini for chat, text-embedding-3-small for embeddings (both via Vercel AI SDK). Streaming at /api/hackbot/stream, non-streaming fallback via askHackbot server action.
• UI: HackbotWidget.tsx fixed-position floating button, localStorage chat history (max 20 messages), 200 char input limit, streaming responses. Widget is mounted in (hackers)/layout.tsx (via HackbotWidgetWrapper) so chat state persists across all hub navigation, and is only visible to authenticated hackers.
• CI/CD: hackbot:seed runs on every deploy (staging and production), embedding events from the events collection and all docs from hackbot_knowledge into hackbot_docs. Admins can also reseed immediately from the /admin/hackbot page without a deploy.