Space

Data Science Initiatives

Big Data pipeline : Lambda Architecture


Task

Design a flexible infrastructure to deploy enterprise-ready datalake on AWS. Maintain it as part of CI/CD pipelines.

Deliverable
Integrating AWS Services

S3, EMR, EKS, Redshift, Kinesis, Athena, Glue, ElasticSearch, SageMaker, Quicksight. With Apache Kafka, Airflow. Kubeflow as a recent addition.



Audience Insight


Need
Product

As a development partner, DSi provides tools and technologies for an integration platform (privately reference-able) that connects hundreds of small financial institutions in the US. The platform offers full-fledge offer management marketing platform, embedding targeted offers within printed/electronic documents, tracking and post-campaign analytics.

Segments

Gender, Age, Location, Interest, Brand-awareness.



Real time fraud detection model


Project Overview

Detecting fraudulent credit card transactions using XGboost and Isolation Forest Algorithm.

Challenge
Steps
  • Feature Exploration - Visualizing and profiling features.
  • Feature Selection - Filtering out features on grounds of high co-relations, constant values, high cardinality, client specifics.
  • Feature Aggregations - Aggregating features based on customer spending behavior i.e transaction amount, frequency, merchant, location, etc.
  • Data Preparation - Excluding co-related aggregated features, encoding and null value handling.
  • Modeling - Predicting and modeling by machine learning algorithms.
  • Deliverable - Hyperparameter tuned XGBoost model and Visualization


  • Churn Prediction


    Status

    Running on production. Binary classification model using xgboost. Deployment through AWS Sagemaker.

    Next Step


    Multi-CDN routing strategy


    Task

    Model routing strategy for multi-CDN environment reflecting on recent Quality-of-Experience factors.

    Multi-Objective

    Reduce latency, improve QoE performance. Constraint of max/min traffic volume commitment, hence control over cost.

    Algorithms

    Linear Programming, Decision Trees, Reinforcement Learning.

    Deployment Environment

    Kubeflow, aws EKS, aws Sagemaker.



    Document Classification


    Product

    DSi is development partner of iControl ESI which is web-based document management and review repository designed for contract attorneys and paralegals in the US. It helps lawyers and law firms organise and sort through millions of text documents to find relevant and productive information.

    Classification Task


    Brand-sentiment analysis (public data real time analysis)


    Need

    A real-time social-media campaign monitoring system to evaluate customer sentiment towards brands by analyzing trending topics and geographic segmentation, thus improving brands’ customer engagement and retention.

    Stack

    Python, Tweepy, Numpy, Pandas, Seaborn, Plotly, Dash, Scala, Kafka, Spark, MySQL, TextBlob, NLTK, Flask, JavaScript.



    Dynamic Solution Innovators Ltd.

    House 177, Lane 2, New DOHS
    Mohakhali, Dhaka 1206, Bangladesh

    Contact Us:

    Dhaka Office: +88017300-374-00
    Telephone: +88-02-9887041
    USA Office: 404-287-0730
    Email:

    Find Us: