Many RAG demos hide the retrieval step and make it hard to tell whether an answer came from source material or model guesswork.
Mini-RAG shows the practical parts of retrieval, grounding, and evals.
This project exists to show that I understand more than a chatbot wrapper: document loading, retrieval quality, answer grounding, guardrails, and simple checks that make a RAG system reviewable.
A small Python demo keeps retrieval, source context, and response checks visible so the failure modes are easy to inspect.
The demo has a public Render service, README, and portfolio links so reviewers can inspect both the running app and implementation.
- Good for showing junior AI engineering fundamentals: chunking, ranking, grounding, and testable behavior.
- Kept intentionally lightweight so the architecture is readable and not buried under a large framework.
- Next upgrade: add a small public corpus switcher and a saved evaluation report for each sample question set.
Render free services can sleep after inactivity, so first load may take around 30 seconds.