pixegami
YouTube creator focused on practical Python tutorials for LLM apps — RAG over PDFs, LangChain, ChromaDB, local model integration, pytest-based AI app testing. Style is implementation-first with full GitHub repos and minimal hype.
Channels
- YouTube: pixegami — Python LLM tutorials, RAG, LangChain, pytest
- GitHub: linked from each video; full reference projects per tutorial
Content in This Wiki
- Python RAG Tutorial (with Local LLMs) — Building a RAG app over PDFs using LangChain + ChromaDB + AWS Bedrock embeddings + Ollama-served Mistral. Includes deterministic chunk IDs for incremental updates and an LLM-as-judge unit testing pattern.
Key Ideas
- Hybrid is fine: cloud embeddings (better quality) + local LLM (cheap inference) is a defensible split for personal RAG apps when local embeddings underperform
- Deterministic chunk IDs (
source:page:chunk_index) let you incrementally add documents to a vector DB without rebuilding from scratch - LLM-as-judge unit tests — pytest cases that ask a separate LLM call whether the actual answer matches the expected answer; include negative cases (with inverted assertions) so you catch over-generous evaluators
- For RAG quality, embedding quality matters more than LLM quality — get embeddings right first