DAVID MUNA MWANGI

Highly adept Data Scientist with close to half a decade of experience in Business Intelligence, AI, Machine Learning and Deep Learning to help organizations achieve their goals of being data-driven through Teamwork, Collaboration, Quality Assurance, and clear communication with stakeholders. Possesses excellent understanding and continuous learning ability of new Data Science techniques. David is currently looking for a Data Science, ML, or AI-related position in a promising environment with more learning and growth opportunities.

• I received an award after the second quarter of 2022 from pOrbis titled, Distinguished Support in AIDA Project using AI and Machine Learning on 1st July 2022 https://www.linkedin.com/posts/porbis-group-inc_awards-porbis-activity-6952471061225492480-7cOv?utm_source=share&utm_medium=member_desktop
• Managed to cross sell credit cards and loan facilities when clients came to my teller booth.
• Best performing Teller in terms of the total number of transactions, turnover, and productivity from August 2015 to February 2016 while at Chase Bank Kenya(IR)

• Python, SQL and R
• Tensorflow
• Statsmodels
• Sklearn/Scikit Learn
• Git and Github
• AWS
• Microsoft Azure
• Docker
• Kubernetes
• VS Code

• Contracted Data Scientist/AI/ML/DL Engineer, Swift Spread Mapping Project for pOrbis Group in Houston Texas, 04/2022 - 07/2022, Worked on a Flask Machine Learning Mapping Project from Ideation phase to deployment into the production environment using Python, Flask, AWS s3 bucket, MongoDB and Postman. Managed to create an LSTM model using tensorflow and other required libraries to automatically upload an excel file or pdf/jpeg/jpg/png, detect information from images of pdf’s such as account numbers, descriptions and amount/credit/debit figures using OCR. The next step was to modify its layout, account configuration and started the account mapping process based on existing mappings, lookup from mappings in a training file in MongoDB collection, and finally mapped using a machine learning model. The mappings were captured from previously saved existing mappings, lookup from training file and the ml model with the confidence, status, allocation percent and source of mapping. Worked on the development environment replication process and tested the api endpoints using Postman for deployment into production. This allowed pOrbis’s dealers to automatically map their trial balance files amount figures to the correct cell in our front end UI based on the account descriptions they provide for the hundreds of thousands of records by OEM Dealers, hence making work easier for them.
• Contracted Data Scientist/AI/ML/DL Engineer, AIDA Time Series Forecasting for pOrbis Group in Houston Texas, 11/2021 - 02/2022, I worked with a dataset with the curse of dimensionality(>3000 features) and hundreds of thousands of records collected in a span of 8 years, to perform analysis and identify trends and relationships between the different features and how they affect outcome variables. My job was to clean the data, reduce the dimensions from >3000 variables to fewer features with high information value and high explained variance through PCA. I performed univariate, bivariate, and multivariate exploratory data analysis to achieve this in addition to Feature Importance Analysis. I identified granger causal relationships of important features across different times and created multiple-output regression models as well as multi-step forecasting models for the variables at different times of the year. This was done by making interactive notebooks that a user can use to filter through and get the desired analysis. I used LSTM Deep Learning models for time series forecasting.
• Data Scientist, Parametric Survival Analysis for Same Day Auto Finance in Texas, Upwork, 09/2021, I performed Parametric Survival Analysis for a client from Upwork using R and obtained a 5-Star rating. I worked with a dataset that had fewer records than features with >35% missing values. My aim was to produce survival curves to estimate how many months it would take for a particular record/client to default on their payments. I used libraries such as smoothsurvreg and survival among others for the analysis and managed to exceed the client’s expectations. Used the weight of evidence technique to deal with missing values and created uniform bins & calculated estimated hazard outputs from day 0 to 48 in the form of a table followed by respective cumulative hazards for each record at month 48. Transposed columns to rows and created equally sized bins based on cumulative hazards and grouped by the bins & compared the bins with the mean of target variable Prepaid. and successfully produced survival curve estimates with an accuracy of > 85%

• Tushar Chaudhary, Data Scientist & ML Engineer, pObis Group LLC, t.chaudhary@porbis.com
• Jedidah Ochieng', Technical Mentor- Moringa School, 0706601058, jedidahakinyi@gmail.com