AI Interactive Tools
Cheat Sheet PDF Generator
Generates personalised A4/Letter cheat sheets for 5 topics (RAG, Prompt Eng, AI Governance, Vector DBs, AI Costs).
TL;DR
One-line answer: Generates A4/Letter PDF cheat sheets for 5 ready AI topics.
- RAG, Prompt Eng, AI Governance, Vector DB, AI Costs.
- TR & EN language; instant in-browser PDF.
- Optional newsletter signup for sharing the PDF by email.
Definition
- Cheat Sheet
- A single-page reference document that summarises the most critical decisions, formulas or commands of a complex topic.
- Also known as: quick reference, crib sheet, kopya kağıdı
Cheat sheet configuration
Topic
RAG Architecture Cheat Sheet
30+ practical decisions to ship a Retrieval-Augmented Generation system to production.
1. Data Layer
- Source: SharePoint, Confluence, S3, DB — pick a connector.
- Chunking: 256-512 tokens, ~15% overlap, keep heading metadata.
- Use dedicated extractor for tables/images (Unstructured.io, Marker).
- PII: mask before indexing (Microsoft Presidio).
2. Embedding & Index
- Pick embedding by MTEB score, test on your language.
- Vector DB: Pinecone (easy), Qdrant (flexible), pgvector (integrated).
- Hybrid: BM25 + vector via RRF or weighted-sum.
- Re-ranker (Cohere Rerank 3, BGE-reranker) lifts accuracy ~15%.
3. Query Pipeline
- Query rewriting: paraphrase user intent with the LLM.
- HyDE: generate a hypothetical answer and retrieve with it.
- Top-k: start with 5-10, tune via evals.
- Citation mandatory: return source URL + paragraph id.
4. Evaluation
- RAGAS: faithfulness, answer relevancy, context precision.
- Golden set: 50+ expert-validated Q&A pairs.
- Hallucination threshold: faithfulness > 0.85.
- Latency: target p95 < 3s, retrieval < 500ms.
5. Production Ops
- Caching: query- and retrieval-level (Redis 1h TTL).
- Rate limit: per-tenant token + RPS quota.
- Logging: prompt, retrieved chunks, response, eval score.
- Fallback: route to model B if A fails; plan refusal flow.
Cheat sheets are summaries; verify with official documentation before production use.