Building RAG with Open-Source Tools
We’re excited to release a brand-new tutorial to allow any AI engineer to build a RAG pipeline from scratch using purely local, open-source modules . No API calls to third-party systems!
sentence_transformers embeddings
Postgres + pgvector as the vector store
Llama 2 as the LLM (through llama.cpp)
Check out our guide here: https://lnkd.in/gfXE5WQ2
The link resolves to a LinkedIn post from LlamaIndex, which points to their documentation tutorial titled "Building RAG from Scratch (Open-source only!)."
Claim: "sentence_transformers embeddings, Postgres + pgvector as the vector store, Llama 2 as the LLM (through llama.cpp)"
Verification against the actual LlamaIndex tutorial: this is accurate. The source confirms the stack uses Sentence Transformers as the embedding model, Postgres (with the pgvector extension required as a separate install) as the vector store, and Llama 2 as the LLM run through llama.cpp. The tutorial also walks through building both an ingestion pipeline and a retrieval pipeline manually, matching the "from scratch" framing in the post.
Claim: "No API calls to third-party systems"
Verification: consistent with the source. All three components (embedding model, vector store, LLM) run locally, with no dependency on hosted third-party APIs.
Minor note: the tutorial does still rely on LlamaIndex's own Python library and abstractions (readers, node parsers, vector store wrappers) to wire the components together. It is not fully "from scratch" in the sense of raw Python with zero framework, though the AI/model components themselves are open-source and local, which matches the post's actual claim.
Verdict: The technical claims in the post are accurate and match the underlying tutorial content.
Note: lnkd.in links cannot be fetched directly due to robots.txt restrictions, so this was verified through the publicly indexed LinkedIn post text and the corresponding LlamaIndex documentation.