My profile image

Hi, I'm Leo Kravtchin!

Data and Business Intelligence Engineer with 4+ years of experience at Amazon, METRO Markets, and several startups. Delivered solutions, lead engineering teams, and managed customers across accounting, risk compliance, and supply chain organisations. Specialised in Python, SQL, and 20+ AWS services for data, software, and GenAI automations. Master’s degree in Computer Science and AI from the University of Edinburgh.


Overview Resume / CV

Here is a quick overview about me


BIE II at Amazon

Business Intelligence Engineer II in Accounting Technologies, working on Python and AWS automation projects.

Master's Degree

Master's degree in Computer Science and Artificial Intelligence from the University of Edinburgh.

GenAI

Building GenAI infrastructure and applications using RAG knowledge bases and MCP servers at Amazon.


Examples of projects and accomplishments

Due to company and university policies, all sourcecode and sensitive details remain private


Thesis result example

Amazon Project Examples

  • Full development lifecycle and on-call workstream agents using MCPs and RAG knowledge bases
  • Increasing development speed by 20% for 70 engineers, delivering accounting automations faster
  • Microservice for scalable Redshift table archival using AWS Glue Catalog across 3 environments

Thesis result example

DelphAi: Amazon ORC Hackathon

In a team of 4, we developed an AWS Lex chatbot using Bedrock, a RAG knowledge base, API Gateway, Amplify, S3, and Lambda to add documents to the RAG model. The chatbot answers questions about Amazon domains, based on documents uploaded by the user through the UI.
AWS Bedrock, API Gateway, Amplify, S3, Lambda, Lex, React, Python, JavaScript

Thesis result example

SlackOracle: Amazon UK&I GenAI Hackathon

In a team of 5, we developed an internal Slack bot that uses AWS Bedrock, a RAG knowledge base, MCP server, API Gateway, Lambda, S3, and OpenSearch to answer users' questions about a Slack channel's history. New Slack channels are ingested into the knowledge base through a Slack bot interface, allowing onboarding to new channels.
AWS Bedrock, API Gateway, Lambda, S3, OpenSearch, Slack

Thesis result example

Master's Thesis

Interpretable and Trustworthy Machine Learning for ICU Admission Length of Stay and Mortality Predictions, supervised by Dr Sohan Seth.
Using machine learning and data science techniques for ICU patients' mortality predictions (98% accuracy) and length of stay predictions (96% accuracy).
Python, TensorFlow, Scikit-Learn, Pandas, Prophet, NeuralProphet, Git

Thesis result example

BSc Thesis

Detecting Long-Term Deviation and External Factors Correlations in Activities of Daily Living Based on Sensor Data, supervised by Dr Jacques Fleuriot.
Using data analysis, machine learning, and deep learning techniques for human activity recognition, clustering, process mining, and activity forecasting, across over 5 years of daily sensor data.
Python, TensorFlow, Scikit-Learn, Pandas, PM4Py, Prophet, NeuralProphet, Git

Metro Markets logo

METRO Markets Projects

  • Python, SQL, and Google Cloud for machine learning, data analysis, and data automations
  • Improved performance monitoring by 50% using JavaScript, Grafana, and Slack bots alerts.
  • Prevented 40k faulty customer orders using SQL and Python analysis during on-call support
  • Created Python data pipelines for diagnostics of internal systems, increasing uptime by 25%
  • Saved 50 hours per week through Power BI dashboards and forecasting reports for leadership
Snippet homepage

Snippet

In a team of six, we developed a full-stack web search engine for 10m Spotify podcast transcript snippets, enabling users to search for and listen to only the parts of podcasts and topics they are interested in. Implemented features include ranked information retrieval, semantic search, topic modelling, query expansion, and CI/CD using Jenkins via GitHub.
Python, Flask, React, PostgreSQL, MongoDB, Docker, Google Cloud, Jenkins

Confusion Matrix

RecognisED

We developed a full-stack mobile app for real-time human activity recognition from wearable sensors in a team of three. We used deep learning to achieve classification accuracies of 94% for 4 activity subsets and 72% for 14 different activities. The app shows the classified activity, step count, and stores historical user data using Firebase.
Python, TensorFlow, Keras, Scikit-Learn, Java, Kotlin, Firebase, Git

GitHub logo

Full-Stack Web Development

  • This website is open-source and uses HTML, CSS, SASS, JavaScript, and jQuery. It is hosted and updated using CI/CD via GitHub Pages.
  • Full-stack web development for Liedfestival Kassel, including a ticket booking system with seat reservations, email confirmations, and a SQL database using the Django web framework.
  • Front-end web development for GetTransfer.com, end-to-end.

Contact Form



Find me on ...