Increase context size

The previous chunk size of 512 tokens appeared to be too limiting for a good match on the prompt.

• adopted the threshold on the embeddings match;
• increased chunk size when parsing PDFs to 1,024 tokens;
• other minor bug fixes.

The model appears to be more responsive when prompted with questions that do match the subject matter of tech articles fed in the Vector DB.