RAG Support Agent / Raahat

Model
llama-3.3-70b-versatile
Embedding
768d
Top-k
5
Threshold
0.35

Ask. Watch it retrieve.

Ask about Raahat — his work, his stack, what he charges. Your question is embedded and matched against the indexed docs. What clears the similarity threshold gets cited. What doesn’t gets an honest no.

Pipeline

embed question → cosine top-k over pgvector → threshold drops weak chunks → cite what survives