First: is your Telugu really Unicode?
Much Telugu in circulation, from government forms to books, was typed in legacy fonts such as Anu, Priyaanka and Kranthi. The page shows Telugu, but the file holds Latin-letter codes, so copied text reads as symbols. Translating that would translate nonsense, so IndicPDF refuses a document set in a legacy font and says so.
The way through is to convert it first. Legacy PDF to DOCX reads each code through the table for the Anu family and writes real Unicode Telugu; the Word file that comes out can then be translated into English.
The English comes back in the same layout
The Word file is edited in place: styles, tables, numbering, header and footer stay as the Telugu document had them. Numbers, dates, section references, amounts and addresses are protected and come back exactly. In the Telugu-to-English run of our test notice, all 6 of 6 were back, and the layout comparison, an element by element check of paragraph, table and page properties, found no difference.
What a round trip shows
Translating our test notice into Telugu and back gave English that read the same as the original except for one name: “Housing Act” returned as “Housing Development Act”. Official names are the thing most worth checking, in either direction. In the check of the Telugu-to-English pair, the Telugu reader found the one sentence tested acceptable. That is one sentence: a reason to read the English, not a measure of quality.
A sentence that cannot be translated
When a Telugu sentence fails a check, it is cut and asked again, then asked as written. A part that still fails stays in Telugu and is highlighted yellow, so a reader of the English sees exactly where the machine gave up.
What it does not take
A scanned page with no text layer is not accepted yet: read it with Telugu OCR first. Footnotes, endnotes and comments are not translated yet. Text inside pictures is not translated.