Machine Learning
Predictive analytics, anomaly detection, automation. Python, TensorFlow, PyTorch for e-commerce, finance, IoT.
Build
AI & Training
Ho Chi Minh City · UTC+7 · Mon–Fri 09:00–18:30
BSP Software Services Corp. builds SOTA systems - models, LLM chat, and computer vision - that companies can run, not just demo.
Mainstream languages, current frameworks, and methods that survive production load.
Predictive analytics, anomaly detection, automation. Python, TensorFlow, PyTorch for e-commerce, finance, IoT.
Chatbots, sentiment, text classification. NLP that reads human language in context.
Fine-tuned chat with RAG for healthcare, banking, insurance, education, customer service.
Hadoop, Spark, and related platforms. Realtime insight wired into AI systems.
Mining hidden patterns for operations, engagement, and product innovation.
Unstructured text, images, multimedia. Retrieval pipelines and training data.
Encryption inside AI systems. Privacy and integrity for finance and healthcare.
Threat detection with pattern recognition and machine learning.
Expert bases for decisions in healthcare, logistics, and legal tech.
Detection, faces, quality inspection. OpenCV, YOLO, deep learning.
Architectures for prediction, language, and images. Tuned for scale.
Hadoop, Spark, Kubernetes for large training jobs and cloud deploy.
Python, Java, C#, Kotlin plus Android, iOS, Flutter, React Native.
Technology chosen per project so performance matches the brief.
Global clients, tight calendars, limited resource - still shipped.
Current tools with security and a path that stays maintainable.
Vision on the line, LLM in the clinic, privacy on the Mac - products already in use.
Automatic counting for food production and packaging. Up to 99.8% accuracy at high speed. Existing cameras. On-prem or cloud. Excel and ERP/MES via Modbus, MQTT, REST. Retrain with a few dozen images.
A webcam becomes an air controller. cvzone + MediaPipe. Virtual mouse, games, robots, healthcare, 3D architecture, virtual cars, machine inspection.
Domain LLM for triage, FAQs, appointments, and service guidance. Fine-tuning and RAG, built to sit inside existing care systems.
MediaPipe FaceMesh. Light enough for existing hardware, open enough to customize for product or research.
OpenCV and deep learning. People, vehicles, animals. Works with cameras already on site - security, farms, homes, plants.
face_recognition + Tkinter. Webcam in/out, leave rules, period reports, SQLite. Built for IT companies.
Counts bicep curls from video or webcam. MediaPipe pose. CLAHE and smoothing against bad light. Ready for mobile or IoT.
Linear regression on Yahoo Finance OHLCV to estimate the close. A clean baseline teams can extend.
“Find the phone.” Camera, clock-face location, spoken Vietnamese. YOLOv5x and CNN.
Encrypt and embed copyright in the image itself. Extract later. For photographers, designers, artists.
Hide encrypted text in JPEG or PNG. No visual change. Offline on the Mac.
One private photo becomes two ordinary camouflage images. Both are required to recover it.
98% class/detect work on the Intel Image Dataset for monitoring and business image pipelines.
Tan@bspai.vn · WhatsApp +84 93 872 8579 · Teams Tan.Nguyen.79