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.
Forwarding letter for FE visa Extension sonTokyo Electric Power Service Co., Ltd.(TEPSCO)
12 fields · learned from 2 examples
Visa Extension (Form II)-Chiaki UchidaTokyo Electric Power Service Co., Ltd.(TEPSCO)
5 fields · learned from 2 examples
Recommendation letterTokyo Electric Power Service Co., Ltd.(TEPSCO)
6 fields · learned from 5 examples
Authorization LetterTokyo Electric Power Service Co., Ltd.(TEPSCO)
3 fields · learned from 4 examples
Forwarding Letter -Mizutani SanTamurakoma & Co., Ltd.
22 fields · learned from 3 examples
Authorization LetterTamurakoma & Co., Ltd.
7 fields · learned from 3 examples
Recommendation letterTAISEI CORPORATION
2 fields · learned from 2 examples
Authorization LetterTAISEI CORPORATION
11 fields · learned from 2 examples
Forwarding LetterTAISEI CORPORATION
6 fields · learned from 2 examples
Authorization LetterNippon Koei Bangladesh
5 fields · learned from 5 examples
Forwarding LetterNippon Koei Bangladesh
11 fields · learned from 6 examples
Authorization LetterKUSUHARA QAMAR NAZ (Bangladesh) Ltd.
7 fields · learned from 3 examples
Forwarding LetterKUSUHARA QAMAR NAZ (Bangladesh) Ltd.
13 fields · learned from 3 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.