Kỹ Sư Phân Tích & Nhận Dạng Tài Liệu (OCR)

Model Development & Optimization

• Maintain and enhance existing AI models for OCR on Vietnamese ID cards (CCCD) and extend to other document types (passports, driver licenses, bank documents).
• Fine-tune and adapt state-of-the-art OCR/document models (Donut) for production use.
• Optimize training and inference pipelines for performance, scalability, and cost efficiency.

Data Pipeline & Quality Management

• Manage large datasets combining synthetic and real-world document images.
• Build preprocessing and augmentation pipelines: image quality checks, blur/rotation detection, Vietnamese text normalization, PII masking.
• Ensure data quality and evaluation consistency across multiple document types.

Accuracy & Performance Evaluation

• Define and monitor evaluation metrics: character/word accuracy, exact match rate, edit distance, latency.
• Analyze failed predictions (e.g., accents, truncated fields, misrecognized entities) and integrate findings into retraining cycles.
• Implement image/document quality control to prevent poor inputs from degrading OCR accuracy.

Production & Monitoring

• Deploy, monitor, and maintain OCR models serving production workloads (100k+ documents/month).
• Investigate and resolve production failures, manage rollbacks, and improve system robustness.
• Collaborate with backend engineers to integrate OCR APIs with downstream systems.

Collaboration & Leadership

• Mentor junior engineers in computer vision and OCR best practices.
• Contribute to the long-term roadmap for Document AI, beyond ID cards, to support broader fintech/eKYC and document processing needs.
• Document experiments, model updates, and operational practices.

Must-have

• 3+ years of AI/ML engineering experience with Python and PyTorch.
• Practical experience in OCR or Computer Vision (e.g., image preprocessing, OpenCV).
• Experience with Vietnamese text processing (accents, tokenization, normalization).
• Familiarity with deep learning model training and fine-tuning, preferably with HuggingFace Transformers or OCR frameworks (PaddleOCR, Tesseract).
• Experience deploying ML models into production environments.
• Experience scaling machine learning services for high traffic.
• Knowledge of Linux, Docker, and Git.

Nice-to-have

• Knowledge of MLOps tools (Weights & Biases, MLflow, DVC).
• Model optimization skills: quantization, distillation, ONNX/TensorRT.
• Background in fintech/eKYC or handling sensitive/PII data.

Soft Skills

• Strong ownership mindset: accountable for the full lifecycle of OCR models.
• Problem-solving ability: capable of debugging training and inference issues.
• Communication skills: explain ML concepts and findings to technical and non-technical stakeholders.
• Collaborative attitude: work closely with backend, product, and QA teams.

Tech Stack

• Python, PyTorch, HuggingFace Transformers, PaddleOCR
• OpenCV, PIL
• Docker, Linux
• Git, DVC (optional)
• MLflow / Weights & Biases (nice-to-have)

Competitive salary package (Base salary and performance bonuses).

Probation period salary is 100% of the official salary.

Comprehensive health and accident insurance.

15 days of annual leave, 3 remote work days per month.

Provision of work equipment (Macbook/ Laptop, mouse, monitor, etc.).

A creative and modern working environment.