Mobile-First Multilingual Translation Model (22 Indian Languages + English)

We are looking for an experienced AI/ML Engineer with expertise in NLP, Machine Translation, and Mobile AI deployment to build a lightweight multilingual translation model optimized for mobile devices. The model must support: 22 Indian languages English Romanized Indian language text Example: "meeru ekkada unnaru" Mixed-language/code-mixed text Example: "Nenu office ki late avuthanu today" The primary objective is to create a high-quality translation system that can run efficiently on mobile devices with low latency and a compact model size. Target Specifications Model Size Preferred: 50MB–100MB Quantized versions encouraged Mobile deployment optimized Performance Goals Fast inference Low memory footprint Offline/on-device support preferred Production-ready deployment Required Capabilities Translation Support The model should support: Indian Language ↔ English Indian Language ↔ Indian Language Native script inputs Romanized Indian language inputs Code-mixed inputs (Indian language + English) Examples: Telugu → English Hindi → Tamil Roman Telugu → English Mixed Telugu + English → Hindi Data Collection & Preparation The selected candidate may use: Open-source datasets Public multilingual corpora Web scraping (where legally permitted) Dataset merging and augmentation techniques Requirements: All data sources must be legally usable Dataset preparation must be reproducible Full documentation required Deliverables Model Deliverables Final trained model Quantized model versions Mobile-optimized model ONNX/TFLite/CoreML export formats Inference-ready deployment package Code Deliverables Training scripts Fine-tuning scripts Evaluation scripts Inference scripts Model export scripts Data Deliverables Dataset collection pipeline Cleaning and preprocessing scripts Tokenizer generation scripts Vocabulary generation scripts Dataset documentation Documentation Complete documentation covering: Environment setup Training process Fine-tuning workflow Reproducibility guide Evaluation methodology Deployment guide Mobile integration instructions Evaluation Requirements Please provide: Translation quality metrics Evaluation methodology Benchmark results Sample translations Latency benchmarks Memory usage benchmarks Model size comparisons The final solution should balance: Translation quality Model size Inference speed Technical Requirements Preferred experience with: Machine Translation (NMT) Multilingual NLP Indic Language Processing PyTorch TensorFlow Transformers Quantization Knowledge Distillation ONNX TFLite CoreML Mobile AI Deployment Experience with Indic language models is highly preferred. Compute Resources Compute resources can be provided if required. Please specify: GPU requirements Training infrastructure recommendation Estimated training time Storage requirements Approximate compute costs Project Milestones Milestone 1 Architecture Design & Technical Proposal Milestone 2 Dataset Collection & Preparation Milestone 3 Baseline Model Development Milestone 4 Training & Fine-Tuning Milestone 5 Model Compression & Optimization Milestone 6 Evaluation & Benchmarking Milestone 7 Mobile Deployment Support Milestone 8 Final Delivery & Documentation Proposal Requirements When applying, please include: Relevant NLP / Translation projects Experience with multilingual models Experience with Indic language NLP Proposed technical approach Expected model size Expected translation quality Estimated timeline GPU/compute requirements GitHub profile Research papers or publications (if available) Previous mobile AI deployment experience Preferred Candidates We are particularly interested in candidates who have worked on: IndicTrans NLLB mBART M2M-100 MarianMT Distilled Transformer models Mobile AI optimization Quantization and compression pipelines Please include examples of similar work and explain how you would approach this project from data collection to mobile deployment.