Abed Matini demonstrates an employee FAQ assistant that converts documents to structured text before retrieval. He compares heading, paragraph, fixed-length and sentence-based chunks, showing how document structure affects citations and answers.
Abed Matini combines keyword and vector search in PostgreSQL and describes ranking results before passing selected context to a small local language model. His example contrasts a fixed retrieval path with an optional tool-using mode.
The demonstration includes local telemetry, consent and simple guardrails for out-of-scope questions. These controls are examples for further testing, not proof that a system is immune to prompt injection or hallucination.
Watch on YouTube
