Hi, I'm Meng Han 👋
Computer Vision Engineer based in Singapore 🇸🇬.
Open to Collaboration!
I'm actively learning edge AI and electronics, and am open to projects, collaborations, or just chatting about AI and technology. Feel free to reach out!
My Journey
Avetics Global
Computer Vision Engineer
- Engineering low-latency vision pipelines for drones on constrained edge hardware.
NUS Office of Internal Audit
AI Engineer Intern
- Built an end-to-end fraud detection system for vendor invoices using OCR, PDF forensics, and document similarity - cut manual audit time by ~30%.
- Designed an explainable unsupervised anomaly detector (Isolation Forest + SHAP) to flag suspicious claims without labeled fraud data.
Changi Airport Group
Machine Learning Engineer Intern
- Boosted taxi demand forecasting accuracy from 48% to 62% at Terminal 3 using XGBoost with time-series feature engineering.
- Automated flight turnaround analysis across 18k+ records to identify ground-crew bottlenecks and reduce departure delays
- Slashed monthly data pipeline runtime from 12h to 8.5h via a SageMaker-based ETL system with enhanced logging.
Cisco-NUS Accelerated Digital Economy Corporate Laboratory
Research Assistant
- Built a real-time human detection and tracking system using YOLOv8 + DeepSORT on overhead fisheye video feeds, achieving 84% mAP at 15 FPS on a standard workstation.
- Developed a low-latency pipeline to convert calibrated video streams into standardised JSON bounding boxes, enabling near real-time motion analysis for digital-twin applications.
Rockship (NUS Overseas College, Vietnam)
AI / Computer Vision Engineer Intern
- Created a RAG system (LangChain + LlamaIndex) that ingests 50+ n8n workflows to auto-generate new workflows
- Developed and deployed robust, scalable LlamaIndex RAG APIs for text and image embedding with built-in OCR safeguards, adopted as boilerplate by internal engineering team to accelerate prototype development
- Improved YOLOv8 nail detection mAP by 20pp via targeted labeling; integrated into an ARKit iOS app for sub-centimeter construction alignment
DSTA (Defence Science and Technology Agency) Land Systems
Robotics Engineer Intern
- Implemented State of the Arts Visual SLAM algorithm RTAB-Map to enable simultaneous mapping and localisation on DJI Robomaster platform for reconnaissance missions.
- Developed a framework to efficiently force-multiply robots by linking them to a central platform using communication protocols such as ROS and MQTT.
Technical Expertise
Languages
- Python (Primary)
- C, C++
- SQL
- Java
- Next.js
ML/AI & Generative Models
- PyTorch
- TensorFlow
- Diffusion Models / VLMs
- NLP, LLM, RAG
- scikit-learn
MLOps & Deployment
- AWS, GCP
- CI/CD, GitHub Actions
- LangSmith, LangFuse
- Docker
- ONNX
- FastAPI