From the source
Perplexity's tutorial walks beginners through building a basic retrieval-augmented generation system in Python, using ChromaDB and two Perplexity models for embedding and answer generation.
- The guide argues that while production RAG is complex, a simple architecture helps newcomers understand how the technique works in practice.
- It breaks the system into three parts: an orchestrator, an information source, and an LLM, and covers indexing and embedding, orchestration, and retrieval and generation.
- The tutorial builds a command-line application that answers questions about a single document, using pplx-embed-v1-0.6b to embed the document before uploading it to ChromaDB and GPT-5.6-Luna, accessed through Perplexity's API, to process retrieved information.
- Perplexity says a paid account is not required, though there will be a small credit cost, and it recommends keeping the API key private.
- The guide also lists best practices: a clean ingestion pipeline, hybrid search, evaluation across accuracy, relevance and latency, and secure handling of private information.