Skip to content

project_development_info.md

Latest commit

 

History

History
40 lines (32 loc) · 2.92 KB

project_development_info.md

File metadata and controls

40 lines (32 loc) · 2.92 KB
Raw

Waypoint: Project Development Information

Project Background & Vision

Waypoint is a project developed under the non-profit Phoenix Tech Solutions. Its primary goal is to mitigate the confusions faced by students, parents, and teachers regarding the ongoing events, rules, and general information of the magnet school FCS Innovation Academy.

Initially built as a first prototype, the project has gained traction and is currently in a collaborative phase with school administrators and the Fulton County school district to deploy this schoolwide AI chatbot and make it a reality for the entire ecosystem.

Core Features & Capabilities

  1. Smart Chat Interface: A real-time, responsive chat system that answers user queries with context-aware information.
  2. RAG-Powered Intelligence: Uses Retrieval-Augmented Generation to parse school documents, perform semantic search, and provide accurate, concise answers.
  3. Staff & Club Directory: Allows users to quickly look up teachers, staff, or clubs.
  4. Timeline View: Visualizes upcoming events and milestones dynamically.
  5. User Accounts: Personalized user experience through secure account management (built with Supabase).

Technical Architecture & Stack

The project is built as a full-stack JavaScript/TypeScript application, utilizing modern web frameworks and AI tools.

Frontend

  • Framework: React 19 and React Router for navigation.
  • Build Tool: Vite.
  • Styling: TailwindCSS 4, enhanced with Framer Motion for animations and tw-animate-css.
  • UI Components: Utilizes Radix UI primitives and Lucide React for iconography.

Backend

  • Environment: Node.js server using Express.
  • Language: TypeScript (run via tsx).
  • Database/Auth: Supabase.
  • Rate Limiting: Custom IP-based request bucket implementation to prevent LLM API spam.

AI & RAG Pipeline

The intelligence of Waypoint is powered by a custom RAG implementation:

  • LLM: Google's Gemini Models (specifically gemini-3-flash-preview via @google/genai) serve as the generative brain.
  • Embeddings: Mistral AI (via @langchain/mistralai) is used to convert document chunks into vector embeddings.
  • Vector Store: A custom in-memory vector store performing cosine similarity search (SimpleMemoryVectorStore extending Langchain's VectorStore).
  • Data Ingestion: Reads structured data from a local data.json, splits it using Langchain's CharacterTextSplitter, embeds it, and serves it to the Gemini model for contextually grounded responses.

Current Project Status

  • Frontend: Core interfaces (Chat, Sidebar, Navbar) and accounts are implemented. Settings pages are in progress.
  • Backend: Data ingestion, retrieval mechanism, and answer generation are fully functioning. Query fine-tuning is currently being refined.
  • Deployment: Running successfully in local development with staging server rollout in progress, leading up to a production rollout.