Who We Are

BJIT established AI division at offshore in April, 2017

3 years of AI experience

Successfully completed 20+ projects using Machine Learning and Deep Learning.

Focus areas are Computer vision and Natural Language Processing (NLP).

3 Years of of AI experience

50+ professional AI engineer

Skilled in machine learning / deep learning languages, frameworks and toolset.

  • Python, C++
  • TensorFlow, Keras, PyTorch
  • AWS SageMaker etc.
50+ professional AI engineers

Clients are Japan & USA company

Panasonic, SourceNext, Jasmy Inc, Dream Online Inc, KJS Inc.

Clients are Japan & USA company

What We Do

Develop intelligent systems using machine learning and deep learning that automates process,
predict, classify objects, engage with human cognitively

Intelligent application development

Cost effective solution with reuse of standard backbone architecture (models and dataset) through transfer learning or solution with new architecture and prepared dataset. Provide robust solution using cloud APIs and SDKs for AI. Provide optimized AI solutions running in both cloud and edge devices.

PoC to verify innovative ideas

Verify feasibility of new research findings or new use cases with PoC in the problem domain of Computer Vision (CV) and Natural Language Processing (NLP).

Our Projects

We have successfully completed 20+ projects using machine learning and deep learning

Level Safety System

Detect level crossing area, person and vehicles near the crossing from IP camera live feed. Identify dangerous situations and play warning.

Achieve real time detection (~15 fps) with up to 100 objects detection in a typical PC.

Both SSD-MobileNet and YOLOv3 based solution are developed. User can select one of the solutions.

Technology

C++, Python, OpenCV, OpenVINO, .NET

Level Safety System
Illegal Parking Identification

Illegal Parking Identification

Automatically identify illegal car parking from video feed. Leverage three different modules to trigger illegal parking. Parking area detection, robust car identification, vehicle tracking and re-identification.

YOLO-v3 for car and parking slot detection, Unet for parking slot segmentation. SORT algorithm for vehicle tracking and Siamense Network is used for vehicle re-identification.

Technology

Tensorflow, Keras, Pytorch, Python, OpenCV and AWS

ASR System

Automatic Speech Recognition (ASR) system - converts English speech into sequence of words using End-to-End Deep Learning.

The system can recognize any clear spoken English and able to adopt with various accent, noisy environments and background sound.

It can transcribe speech in 5% Character Error Rate (CER) for American accent and 15% CER for Philippine accent.

Technology

Tensorflow, Pytorch, Python, Librosa, C#

ASR System
Web Filtering System

Web Filtering System

Categorize web content to filter out adult, crime, hate web-sites. System learns to classify URLs into different categories using Deep Learning.

DMOZ training dataset is used with 3.5 millions of URLs. Categories are Adult, Sports, Games, Social Networking, Dating, Movies, Music, Cartoon/Anime, Comics, Suicide, Shopping, Crime, Gambling.

Accuracy is 95.34% on the test set. Test set has 33,037 samples.

Technology

Tensorflow, Keras, Python

Voice Activity Detection

This goal is to develop a Voice activity detection model that works in real time. Any sound rather than voice is identified as Noise.

Model is trained using convolutional neural network with log mel-spectrogram features.

Classification model is deployed on Android using Tensorflow Lite.

Technology

Tensorflow Lite, Keras, Python, Librosa, Android

Voice Activity Detection
Smart Door Lock System

Smart Door Lock System

Smart door lock system uses face recognition for screening individuals. Four categories: Family members, Delivery person, Stranger (Criminal), Visitor are supported.

Used pre-trained Inception ResNet v1 model trained on VGGFace2 dataset.

Technology

Tensorflow, Python, Facenet pre-trained model, OpenCV for face detection

Survey Chatbot

Chatbot interviews users about a product experience. It extracts information from the user by asking questions.

Machine Learning is used to check if user’s answer is appropriate for the asked question. Extract product related feedback from user by following question hierarchy.

Technology

Tensorflow, Keras, Python, MeCab morphological analysis engine, fastText pre-trained word embedding

Survey Chatbot