Tamil Newspaper Translation Model

I need a skilled AI engineer to build a custom machine-translation model that converts English and Hindi news copy into Tamil while reproducing our national daily’s distinctive tone, layout cues, and editorial policy. We have no legacy system, so the work begins at green-field stage: data gathering, model architecture, training, evaluation, and a lightweight delivery pipeline are all part of the brief. The engine must handle straight news reports and opinion pieces, preserving our house style—headlines, sub-heads, paragraph length, punctuation preferences, and the restrained, formal vocabulary our readers expect. Output should be ready to drop into our CMS with minimal editing. Newspaper pdf pages will be given to know the style and format and to train the model and also the corpus of vocabulary words which are being used by the newspaper will be given. Deliverables • A trained neural model (checkpoint files and inference script) that ingests raw English or Hindi text and returns polished Tamil in our format • A preprocessing pipeline for tokenisation, transliteration where necessary, and style tagging • Deployment code or API endpoint, Dockerised for our servers • A concise user manual covering setup, typical workflows, and fine-tuning steps • Benchmarks demonstrating translation quality (BLEU or equivalent) and style adherence on a held-out set of news articles and editorials Completion is confirmed when the model scores at or above the agreed benchmark and our editorial desk signs off on a live test batch.

ID del proyecto: 39754441