Building Online RAG Pipelines For Semantic Search — Data skill for Claude Code

Building Online RAG Pipelines for Semantic Search

A complete, hand-written Retrieval-Augmented Generation (RAG) pipeline in plain JavaScript — seven steps, zero frameworks, fully functional offline by default.

**Course repository** for the 51-slide "Building Online RAG Pipelines for Semantic Search" (2026).

---

Quick Links

Resource	Location
Course materials	`rag-pipeline-course/`
Full documentation	`rag-pipeline-course/docs/`
Wiki (concepts & guides)	GitHub Wiki
Slide-to-code mapping	`rag-pipeline-course/docs/slide-to-code-map.md`

---

The Pipeline (7 Steps)

Data → Clean → Chunk → Embed → Store → Retrieve → Rerank → Prompt → Generate
   1      2       3       4        5        6        7        8       9

• Offline (steps 1-4): Run on a schedule — ingestion, chunking, embedding, indexing
• Online (steps 5-9): Run per-request — semantic search, reranking, grounding, generation, streaming

---

Run It (No Keys Required)

git clone https://github.com/etemigarba/Building-Online-RAG-Pipelines-for-Semantic-Search.git
cd Building-Online-RAG-Pipelines-for-Semantic-Search/rag-pipeline-course
npm install
npm run index:seed    # index 12 seed documents (~40 chunks)
npm run serve         # http://localhost:3001

Ask: *"What is the pass mark for CSC 508?"*

Uses deterministic fake providers by default (`EMBED_PROVIDER=fake`, `LLM_PROVIDER=fake`) — no API keys, no network, no Docker.

---

Go Live (Two Variables)

EMBED_PROVIDER=openai OPENAI_API_KEY=sk-... \
LLM_PROVIDER=claude   ANTHROPIC_API_KEY=sk-ant-... \
npm run index -- --full && npm run serve

Add PostgreSQL/pgvector when the JSON store outgrows linear scan:

createdb ragdb && psql ragdb -c 'CR