Skip to content

score.md

Latest commit

 

History

History
39 lines (29 loc) · 2.29 KB

score.md

File metadata and controls

39 lines (29 loc) · 2.29 KB

Scoring: Knowledge Hub RAG & Vector Database

Max total score: 194

Basic Scope

  • +20 POST /ai/rag/index indexes Knowledge Hub article data into vector storage correctly
  • +16 POST /ai/rag/search performs semantic search and returns ranked chunks with article attribution
  • +26 POST /ai/rag/chat implements end-to-end RAG (retrieve relevant chunks, build grounded prompt, return answer + sources)
  • +10 DELETE /ai/rag/index/articles/:articleId removes article vectors from index correctly
  • +16 Gemini generation + Gemini embeddings are integrated correctly (configured via .env)
  • +16 Chunking configuration is implemented via env (RAG_CHUNK_SIZE, RAG_CHUNK_OVERLAP)

Advanced Scope

  • +20 External vector DB is used via Docker Compose (separate container, healthy startup, app connectivity)
  • +10 Retrieval supports metadata-aware filtering (status/category/tags)
  • +10 Conversation memory for RAG chat is implemented with configurable max history size
  • +10 Reindex strategy handles updates/deletes consistently (no stale vectors in results)
  • +10 Resilient error handling for vector DB/Gemini outages (503 + logs)

Hacker Scope

  • +10 Hybrid retrieval (semantic + lexical) with merged ranking
  • +10 Secondary re-ranking strategy improves relevance of top chunks
  • +10 Incremental indexing pipeline (index only changed articles, with idempotent behavior)

Forfeits

  • -95% of total task score Any external tools/libraries beyond those listed in technical requirements
  • -30% of total task score Commits after deadline (except commits that affect only Readme.md, .gitignore, etc.)
  • -20 Missing PR or its description is incorrect
  • -20 No separate development branch
  • -20 Less than 3 commits in the development branch, not including commits that make changes only to Readme.md or similar files (tsconfig.json, .gitignore, .prettierrc.json, etc.)
  • -15 RAG remains in-memory only and does not use external vector DB container
  • -10 Gemini integration is missing or replaced with a different LLM provider
  • -10 README does not contain complete runbook (Gemini setup + vector DB startup + indexing flow)
  • -5 The .env file with actual API key is present in the repository (should be .env.example instead)