Everything here is rule-based: values are substituted into your own approved file, and every finding is proved from the text itself. The same input always gives the same result.
What it has learned here, and teaching it a new type from approved examples.
Draft] Retrenchment NoticeFast Value Global (BD) Ltd
8 fields · learned from 9 examples
Letter of GuaranteeFast Value Global (BD) Ltd
9 fields · learned from 9 examples
Letter of Guarantee (Angikarnama in English) ofFast Value Global (BD) Ltd
9 fields · learned from 2 examples
AngikarnamaFast Value Global (BD) Ltd
9 fields · learned from 25 examples
Final calculationFast Value Global (BD) Ltd
44 fields · learned from 16 examples
Termination Letter ofFast Value Global (BD) Ltd
4 fields · learned from 10 examples
Termination Letter bangla ofFast Value Global (BD) Ltd
2 fields · learned from 8 examples
Pick two or more approved documents of the same kind that differ only in their details — three notices for different shareholders, say. Comparing them shows which parts are the format and which are the facts. The first one you tick is the base, and every draft is built from that exact file, so drafts keep its letterhead and layout.