feat: add RAG model for project search and recommendations (closes #2)

[misakaAron]

Summary

This PR implements a Retrieval-Augmented Generation (RAG) model for project search and recommendations, as requested in #2.

Changes

1. Database Schema (prisma/schema.prisma)

• Added embedding field (pgvector 768-dim) to the Project model for storing vector embeddings

2. Embedding Utilities (lib/embedding.ts)

• getEmbedding(): Generates embeddings using Google's ext-embedding-004 model
• embeddingToPgVector(): Converts embedding arrays to PostgreSQL vector format
• projectToText(): Builds a text representation of a project for embedding generation

3. RAG Core (lib/rag.ts)

agSearch(): Performs semantic search over projects using cosine similarity via pgvector

• updateProjectEmbedding(): Generates and stores embedding for a single project

ebuildAllEmbeddings(): Batch rebuilds embeddings for all projects

4. API Routes

• GET /api/rag/search?q=: Semantic search endpoint returning similar projects with similarity scores
• GET /api/rag/recommend?projectId=: Returns projects similar to a given project
• POST /api/rag/recommend: Triggers embedding generation (single project or all)

5. Chat Integration (�i/actions.ts)

• Updated generateRecommendations() to first retrieve semantically similar projects from the database (RAG retrieval step), then use those as context for the LLM to generate better recommendations

Requirements

• Requires pgvector extension enabled on the PostgreSQL database
• Requires GOOGLE_GENERATIVE_AI_API_KEY environment variable

[vercel]

@misakaAron is attempting to deploy a commit to the fahad-dezloper's projects Team on Vercel.

A member of the Team first needs to authorize it.