Looking for Expert NLP/ML Engineer for Language Translation Model Training (Indic Languages)

Project Description: I am looking to hire an experienced NLP/ML engineer to train high-quality machine translation models for Indic languages. The goal is to develop single language-pair models, such as: ● English → Telugu ● English → Hindi (and additional language pairs, if needed) You may choose the most suitable model architecture based on your expertise (e.g., mBART, mT5, NLLB fine-tuning, Transformer variants, etc.), as long as the final models deliver strong translation quality. Dataset: ● You can use the AI4Bharat datasets including: ● Samanantar ● BPCC ● Other open Indic parallel corpora Scope of Work: The freelancer will be responsible for: 1. Data Handling ● Cleaning, filtering, and preprocessing datasets Sentence alignment (if needed) ● Tokenization and vocabulary preparation (SentencePiece/BPE/etc.) 2. Model Training ● Selecting an appropriate model architecture ● Training single language-pair translation models ● Implementing best practices for training efficiency (FP16, gradient accumulation, etc.) ● Hyperparameter tuning Checkpoint management and monitoring 3. Evaluation ● Compute BLEU, SacreBLEU, and other relevant metrics ● Provide side-by-side qualitative translation samples ● Benchmarking against baseline models 4. Delivery ● Final trained model weights ● Inference scripts (Python) for quick testing ● Instructions for running and continuing training ● Documentation of preprocessing and training pipeline ● Optional: Dockerfile or virtual environment setup Requirements: The ideal candidate should have: ● Strong experience in NLP, Transformers, and neural MT models ● Prior work with Indic languages (big plus) ● Experience with training libraries such as PyTorch, Hugging Face Transformers, Fairseq, OpenNMT, or similar ● Ability to handle large-scale training and dataset preprocessing ● Familiarity with SentencePiece, tokenization strategies, and MT evaluation metrics ● Ability to deliver clean, well-documented code Additional Notes: ● Compute resources can be discussed (I can provide compute, or you can use yours). ● More language pairs may be added later as separate follow-up projects. ● Quality of translation is the highest priority. Apply tot his job