$$
{{ $t($store.state.user.experience_value_in_dollars) }}
Junior
{{ $t($store.state.user.experience_search_name) }}
0
jobs
Machine Learning practitioner obsessed with learning and applying new technologies to create measurable value
Daniel Safai Consultant
,
København, Denmark
Experience
Other titles
Skills
I'm offering
Statistics, Econometrics and Machine Learning specialist with a keen interest in building software solutions and process automation in order to ensure ongoing value for clients and customers.
Daniel’s most significant professional achievements includes successful implementation of statistical models improving products and decision making at major Danish and international organizations such as Danske Bank, Deloitte, SDC, http://********.**, Pandora, Topdanmark, Novo Nordisk amongst others. He furthermore holds a gold medal form the world championship in Econometrics.
Daniel has gained an in-depth understanding of how to create business value from data, having both academic and professional experience with state-of-the-art quantitative techniques and how to use these techniques to create measurable value through data-driven decision making.
While not working on commercial ideas, Daniel is a competition contributor on the Data Science & Machine Learning community Kaggle, and extremely interested in learning and experimenting with new technologies in general. Furthermore, he is also engaged in several blockchain technologies primarily focusing on Ethereum and related decentralized applications.
An outgoing and approachable character that enjoys bouncing ideas around to give the best possible insight into how statistics and especially machine learning can be applied in each specific client case.
Daniel’s most significant professional achievements includes successful implementation of statistical models improving products and decision making at major Danish and international organizations such as Danske Bank, Deloitte, SDC, http://********.**, Pandora, Topdanmark, Novo Nordisk amongst others. He furthermore holds a gold medal form the world championship in Econometrics.
Daniel has gained an in-depth understanding of how to create business value from data, having both academic and professional experience with state-of-the-art quantitative techniques and how to use these techniques to create measurable value through data-driven decision making.
While not working on commercial ideas, Daniel is a competition contributor on the Data Science & Machine Learning community Kaggle, and extremely interested in learning and experimenting with new technologies in general. Furthermore, he is also engaged in several blockchain technologies primarily focusing on Ethereum and related decentralized applications.
An outgoing and approachable character that enjoys bouncing ideas around to give the best possible insight into how statistics and especially machine learning can be applied in each specific client case.
Markets
Denmark
Norway
Links for more
Once you have created a company account and a job, you can access the profiles links.
Language
Danish
Fluently
English
Fluently
Ready for
Larger project
Ongoing relation / part-time
Full time contractor
Available
My experience
2018 - 2018
freelance
Using statistical modelling and machine learning to recommend financial products
The project was an experiment testing the potential for using Machine Learning to improve sales recommendations. The project consisted of creating a recommendation engine that could predict which products existing customers would use in the next month based on past behaviour and behaviour of similar customers.
Tools and technologies used:
• Deep learning – neural networks
• Standard natural language preprocessing
• Python (keras, scikit-learn, numpy, pandas, NLTK, spacy)
Project results:
The project showed promising results, leading to a second project focusing on implementation of the model. Ultimately, the implemented model will provide sales input to sales personnel when in contact with clients.
Tools and technologies used:
• Deep learning – neural networks
• Standard natural language preprocessing
• Python (keras, scikit-learn, numpy, pandas, NLTK, spacy)
Project results:
The project showed promising results, leading to a second project focusing on implementation of the model. Ultimately, the implemented model will provide sales input to sales personnel when in contact with clients.
Python, Machine learning, Deep learning, Statistics
2015 - 2017
job
Senior Consultant
Deloitte Consulting.
Working with data and quantitative analytics to create profound new insights and opportunities for businesses.
Helping decision makers in large cap companies on how to make better decisions using data for e.g. choosing optimal store locations, identifying adversarial behavior, targeting customers etc.
Helping decision makers in large cap companies on how to make better decisions using data for e.g. choosing optimal store locations, identifying adversarial behavior, targeting customers etc.
Advanced Analytics, Machine learning, Deep learning, Python, Keras, Scikitlearn, Statistics
2017 - 2017
freelance
Using deep neural networks to automate decision making and improve customer experience
The goal of this project was to predict the expected insurance claim from damaged cars. By instantly being able to automatically predict insurance claims, customers could get immediately response when an accident occurs. Data consisted of all historical records of damaged cars and the corresponding insurance report. The solution was first to construct an image preprocessing algorithm combining each image to one single representative image of the damaged car. Based on the processed images a deep neural network was trained to fit the result of the historical cases.
Tools and technologies used:
• Python (keras, scikit-learn, numpy, pandas, opencv, PIL)
• Amazon web services
Project results
The most promising part of the model was being able to categorize damages leaving the car unrepairable from car damages requiring thorough human inspection. The model saves the human effort of inspecting and verifying unrepairable cars - a task requiring several full-time employees. The model is being implemented as part of a customer app making insurance easier and more fun for the customer and improving the overall experience.
Tools and technologies used:
• Python (keras, scikit-learn, numpy, pandas, opencv, PIL)
• Amazon web services
Project results
The most promising part of the model was being able to categorize damages leaving the car unrepairable from car damages requiring thorough human inspection. The model saves the human effort of inspecting and verifying unrepairable cars - a task requiring several full-time employees. The model is being implemented as part of a customer app making insurance easier and more fun for the customer and improving the overall experience.
Python, Machine learning, Statistics
2016 - 2017
freelance
Using machine-learning to beat experienced human accuracy
The project involved processing house condition and electricity reports in order to determine whether an insurance could be obtained for a given house. Traditionally, a team of assessors read the reports thoroughly and based on these determine whether an insurance is given and with which precautions/reasons. The goal was to automate this process and validate to which degree a full automation would be possible. Due to the surprisingly large degree of conflicting historical results, test results from automation showed better performance.
The solution involved two approaches to data - structured and unstructured. In the structured approach, the reports were parsed for useful information based on domain specific knowledge from the assessors. The unstructured approach utilized pre-trained word embeddings to quantify the reports and feeding these to a deep neural network.
Tools and technologies used:
• Python (keras, scikit-learn, numpy, pandas, NLTK, spacy)
• R (xgboost, mxnet, dplyr, SnowballC, stringr, tm)
• Tableau
• Amazon warehouse services
Project results
The full implementation of the resulting model is going through an A/B testing period to validate its performance in a fully automated production environment. Part of the model is being used as a tool helping assessors save up to half their work load while ensuring consistency between the assessors’ decisions to avoid conflicting results.
The solution involved two approaches to data - structured and unstructured. In the structured approach, the reports were parsed for useful information based on domain specific knowledge from the assessors. The unstructured approach utilized pre-trained word embeddings to quantify the reports and feeding these to a deep neural network.
Tools and technologies used:
• Python (keras, scikit-learn, numpy, pandas, NLTK, spacy)
• R (xgboost, mxnet, dplyr, SnowballC, stringr, tm)
• Tableau
• Amazon warehouse services
Project results
The full implementation of the resulting model is going through an A/B testing period to validate its performance in a fully automated production environment. Part of the model is being used as a tool helping assessors save up to half their work load while ensuring consistency between the assessors’ decisions to avoid conflicting results.
Python, Machine learning, Deep learning, Statistics
2013 - 2014
job
Teaching Assistant & Supervisor
Københavns Universitet.
• Lectured the master/PhD course Advanced Microeconometrics
• Supervised students writing seminar papers in advanced microeconometrics.
• Supervised students writing seminar papers in advanced microeconometrics.
Statistics
My education
2009
-
2015
Københavns Universitet
Cand.polit, Statistics, Econometrics and programming
Cand.polit, Statistics, Econometrics and programming
• GPA is 11.0 and within the top 5 % (based on graduates in 2014)
• Winner of Econometric Games 2014 – The World Championship of Econometrics
• Publication: “Predicting Poverty With Incomplete Data: Tackling Measurement-Error With a Pseudo Copula Approach”. TopQuants Newsletter, Vol. 2, nr. 2.
Daniel's reviews
Daniel has not received any reviews on Worksome.
Contact Daniel Safai Consultant
Worksome removes the expensive intermediaries and gives you direct contact with relevant talent.
Create a login and get the opportunity to write to Daniel directly in Worksome.
are ready to help you
and get specific bids from skilled talent in Denmark