Machine Learning Engineer

We are a Singapore-based technology company building for parts of the economy the industry has long overlooked. We bring serious engineering — including modern AI and machine learning — to problems most companies have treated as too hard, too small, or too messy to be worth it. Zillwork is one of these. It focuses on a vast and underserved part of the global economy — a workforce that is in demand yet poorly served by the systems built around it. Zillwork is building the infrastructure to change that. The problems we've taken on are genuinely hard, the impact is real, and the people we're building for have been overlooked for too long. If that's the kind of work you've been looking for, we'd like to meet you. Own the model lifecycle: training, fine-tuning, evaluation, and compression for inference on low-end hardware.

You'll own

• Server-side training/fine-tuning (PyTorch / TensorFlow) on GPU; distributed training and mixed precision.
• Model compression for constrained devices:

quantization (INT8/4), pruning, knowledge distillation to a small model; export to a cross-platform

on-device inference runtime .

• Transfer learning / fine-tuning of transformer and speech models (ASR/TTS) for an Indic language.
• Evaluation harnesses, experiment tracking (MLflow/W&B-class), and model versioning/registry.
• An MLOps training loop: reproducible pipelines, data/version lineage, retraining triggers.

Must-have

• 4+ yrs ML engineering; expert Python, PyTorch or TensorFlow.
• Models trained/fine-tuned to

production , not experiments.

• Hands-on

quantization / distillation / pruning and edge/mobile export.

• Transformer architectures; embeddings; evaluation methodology.

Strong signals

• On-device/edge inference; ASR/TTS or low-resource-language modeling; distributed/multi-GPU training; CUDA/inference-optimization; RL.

Auto-disqualify : research-only, nothing deployed · no compression/edge experience · "trains models on phones" misconception · AI-generated application.