What's inside — the complete RAG ship kit
- Weekend build plan — realistic Sat/Sun schedule from ingest to shipped, with checkpoints. No guesswork.
- 3 starter templates — minimal RAG in ~80 lines (Python), production RAG with re-ranking + citations, Next.js RAG API route. Copy, configure, deploy.
- Chunking & embeddings deep-dive — sizes, overlap, semantic vs fixed, which models to pick, metadata to always store. Avoid the #1 RAG failure mode.
- Retrieval tuning that works — top-k, MMR, hybrid BM25+vector, re-ranking, query rewriting, killing hallucinations with grounding. Battle-tested configs.
- Evaluation harness — measure retrieval recall, faithfulness, answer relevance with a repeatable script. Ship with confidence, not hope.
- Ship & monetize checklist — cost controls, latency, guardrails, and 5 ways to sell RAG apps as a service (agency builds, SaaS, white-label, consulting, template sales).
Indie devs Ship a real RAG product over one weekend — not months.
Agencies Sell "docs Q&A / knowledge bot" builds at £2K+.
SaaS teams Add grounded AI answers to your app this sprint.
Consultants Deploy internal knowledge bots for clients in days.
Preview — actual retrieval function & eval harness inside
This is a real sample from the kit — the Python retrieval function and eval harness. Every template is this complete.
Python: RAG Retrieval Function
def retrieve(query: str, k: int = 5):
"""Embed query, search vector DB,
return top-k chunks with scores."""
q_emb = embedder.encode([query])
scores, idxs = index.search(q_emb, k)
chunks = [documents[i] for i in idxs[0]]
return [
{"text": c, "score": float(s)}
for c, s in zip(chunks, scores[0])
]
def build_prompt(query, chunks):
context = "\n---\n".join(
c["text"] for c in chunks
)
return f"""Answer using only this context.
If unsure, say you don't know.
Context:
{context}
Question: {query}"""
Eval Harness (sample queries):
Q: "What's our refund policy?"
Expected: mentions 7-day money-back
Score: 0.94
Q: "How do I integrate the API?"
Expected: references /api/products
Score: 0.87
+ 3 runnable templates (Python + Next.js), tuning guide, and full eval suite.
£35
One-time. Instant access. Lifetime updates.
Built by operators shipping AI apps in production.