Oct 2021 – Present · 4 yrs 7 mos · Remote

• Built a solution to optimize back-of-house workforce in Carrefour hypermarkets in UAE
• Overhaul of a solution to optimize checkout counters workforce in Carrefour stores across 5 markets
• Upgraded 2 year old Airflow (1.10.11 -> 2.7.0) and Superset (0.36 -> 3.0.1) instances
• hosted on EC2. Afterwards, I worked with the DevOps team to migrate them to EKS
• ensuring seamless user transition.
• Developed a chat bot / AI assistant for personalized recommendations using Chainlit, Langchain,
• Azure Open AI services by employing RAG based methodology.
• Developed an internal AI Academy course to equip colleagues with the skills to build AI assistants with RAG based methodology
• Developed backend for personalisation on the SHARE Rewards app (1 million+ downloads), which serves dynamic personalised content based on users transactions, followed brands, interests and in-app activity

Mar 2021 – Present · 5 yrs 2 mos · Kaunas, Kauno, Lithuania

• I'm a mentor for students interested in data analytics related studies in SKILLed FinTech programme. Currently mentoring 3 students, which are in different phases of their university journey. Giving out advice on career choices, job search, sharing my technical experience
• https://students.ktu.edu/people/marius-vileiniskis/
• https://studentams.ktu.edu/people/marius-vileiniskis/

Oct 2020 – Oct 2021 · 1 yr · Vilnius, Vilniaus, Lithuania · Hybrid

• Enabling sales organisation to understand infrastructure costs to run Mambu SaaS cloud banking platform in a cloud agnostic setup (GCP, AWS, Azure, Oracle) to maintain the desired sales margin when individually pricing offers to the clients
• Established team workflows using GCP services including Dataflow, BigQuery, Cloud Composer and Vertex AI to collect, analyse, optimise and predict cloud cost data in cloud agnostic setup

Jan 2019 – Oct 2020 · 1 yr 9 mos · London, United Kingdom

• Client facing role with a big technical emphasis
• Predicting anomalies in offshore electric cables:
• Prevent outages of offshore windfarms by enabling the engineers to perform predictive maintenance on electric cables buried on the seabed by detecting the degrading state from acoustic and temperature readings
• Predicting the no-show passengers:
• Enabling the team to perform Big Data analysis using Watson Studio and Spark to handle large data processing pipelines. The challenge was to move from in-memory computing with Python to a highly parallelised solution with Spark
• Detecting fraudulent credit card transactions:
• Helping the client to prevent both financial and credibility losses due to the fraudulent transactions being validated as non-suspicious and thus authorised. In order to achieve this a machine learning model to detect
• whether a transaction is fraudulent or not for card not present transactions was built
• Data Science Platform on IBM Cloud:
• Developing a data science platform based on IBM Cloud services including Watson Studio, Watson Machine Learning, Analytic Engine, Kubernetes and OpenScale. Establishing the best practices of delivering data science projects using the built platform by clearly separating the roles and responsibilities of data science, developer and operations teams
• AI fairness and explainability capability demonstration of Watson OpenScale:
• An extensive PoC to demonstrate existing capabilities in Watson OpenScale for explainability, bias detection, data drift and other topics on a variety of common use cases in financial sector
• Anomaly detection in VoIP testing framework:
• Design and implement an anomaly detection process with most likely root causes for VoIP call tests based on the packet captures (PCAP) files. Building a dashboard with Dash in order to allow test automation engineers to analyse historical root causes, debug each test ran on a system on an application layer (SIP messages) with helpful visuals and proposed root causes

Oct 2018 – Dec 2018 · 2 mos · Vilnius

Sep 2016 – Oct 2018 · 2 yrs 1 mo · Vilnius, Vilniaus, Lithuania

• Projects:
• Fraud detection for international payments:
• Developing machine learning models to extend the rule based fraud detection solution
• AML adverse media screening tool:
• NLP + machine learning to screen new bank customers for adverse media on money laundering. Project has been a runner up for most innovative automation project in Shared Services & Outsourcing Excellence Awards
• Collateral management email automation:
• NLP + machine learning to automate collateral processing
• Risk score modelling experimentation system on PySpark + Django
• A tool that generates PySpark code from an exported FICO software rule model to expedite the simulation time of tuning rule thresholds up to 20x

Oct 2014 – Aug 2016 · 1 yr 10 mos

• Industrial project: A People Centred Approach to Intelligent, Proactive, Predictive (PCIPP) asset management using Remote Condition Monitoring (RCM) data

Oct 2011 – Dec 2015 · 4 yrs 2 mos

• PhD project in Fault diagnostics for engineering systems
• The main aim of this thesis is to investigate available techniques and develop a methodology for the fault detection and diagnostics for two engineering systems, namely railway point systems and three-phase separators. The two systems represent two different situations occurring when dealing with fault detection and diagnostics. The first system has only one observable sensor which represents the operation of the system, while for the second system several sensors are available to monitor the operation of the system.
• The fault detection of the railway point systems (RPS) was performed on the measured current from the motor of point operating equipment (POE). A threshold based alarm technique is commonly used by railway infrastructure operators. The inability to timely detect the incipient faults of the RPS is a big deficiency of such a technique. The method of One Class Support Vector Machines (OCSVM) together with some elastic metrics has been proposed to overcome this weakness of the threshold based alarm technique.
• The fault detection of three-phase separators (TPS) on chemical process plant was performed given the sensor readings of flow and level transmitters of TPS. A threshold based alarm technique is commonly used by oil and gas infrastructure operators to monitor the operation of TPS. The late detection of faults of the TPS is a big deficiency of such a technique, since it causes the oil and gas processing plants to be shut down. The method of Bayesian Belief Networks (BBN) has been proposed to overcome this weakness of the threshold based alarm technique.

Feb 2009 – Jul 2010 · 1 yr 5 mos

• Statistical programming with SAS, data management with EpiData and ClinTrial, maintenance of web page content

Dec 2008 – Jun 2009 · 6 mos

• Maintenance of computer network, maintenance of web page