Machine Learning Engineer, LLM/VLM - Monee
Experienced (Individual Contributor) Singapore
About Monee
Monee is a leading digital payments and financial services provider in Southeast Asia, with a growing presence in Latin America. Its mission is to better the lives of individuals and businesses in the region with financial services through technology.
Monee delivers a comprehensive range of digital financial products and services, including mobile wallets, payment processing, credit, banking, and insurtech. By leveraging technology and strategic partnerships, Monee aims to drive greater digital financial inclusion, convenience, and accessibility for individuals and small businesses across Southeast Asia and Latin America.
Monee is part of Sea Limited, a global technology company.
Job Description
- Build key information extraction (KIE) models for financial documents (bank slips, payslips, tax forms, statements, ID documents) across SEA markets using open-weight vision-language models (VLMs), covering data construction and labelling pipelines, fine-tuning, evaluation, deployment and regression testing.
- Develop domain-adaptive post-training for open-source VLMs (continued pre-training, SFT, meta / in-context learning), building high-quality multilingual document and replay datasets.
- Improve extraction performance with post-training methods, including reinforcement learning (DPO, GRPO, PPO / RLHF), reward-model design and verifiable rewards for instruction following, reasoning and faithful extraction.
- Build verifier models that judge the truthfulness, faithfulness and groundedness of extracted fields, produce calibrated confidence, and flag uncertain or inconsistent documents.
- Develop document fraud detection, e.g. detecting edited, generated or template-reused documents with image forensics, similarity and graph-based methods over document structure, or deepfake-detection techniques.
- Design, tune and evaluate system prompts and prompt structures, few-shot and retrieval-augmented prompting, and prompt-optimisation workflows for extraction and review tasks.
- Build explainable policy models that learn decision rules from human reviewer decisions, automate rule discovery, and surface inconsistencies in review standards.
- Drive continuous optimization based on offline evaluation, user experience, latency, cost, and stability.
- Collaborate with Risk, Product, Operations and Engineering to turn business problems into modelling tasks and evaluation metrics, and contribute to reproducibility standards and internal technical documentation.
Requirements
- Bachelor's degree in Computer Science, Mathematics, Statistics, or a related quantitative field.
- Proficiency in Python and a deep learning framework (PyTorch preferred).
- Solid understanding of neural network architectures, including Transformers, sequential models, embedding-based models and vision-language models.
- Hands-on project, thesis or internship experience with at least one of: LLM / VLM fine-tuning or post-training (SFT, DPO, RLHF, GRPO), large-scale pre-training, meta learning or in-context learning.
- Familiarity with LLM prompting techniques, prompt structure and prompt optimisation.
- Familiarity with model evaluation: building test sets, defining metrics, error analysis and comparing models fairly.
- Strong communication skills to collaborate with cross-functional stakeholders across Risk, Product, Operations and Engineering.
- Self-driven, comfortable reading papers and model cards and reproducing results from open-source code.
Nice to Have
- Experience with inference, efficiency and performance optimisation (vLLM, quantisation, batching, serving on a single GPU).
- Experience with multi-GPU or distributed training, and parameter-efficient methods (DeepSpeed, FSDP, LoRA / QLoRA at scale).
- Experience with RAG, embeddings and vector databases.
- Exposure to agentic workflows and tool-using LLM systems.
- Exposure to document or image forgery detection, deepfake detection, or graph learning.
- Exposure to classical NLP and OCR techniques (tokenisation, sequence labelling, text post-processing, OCR engines).
- Experience in fintech, credit risk, or financial services.