Skip to content
Portrait of Andrei-Flavius Văcaru

I build machine learning that holds up outside the notebook.

Data scientist. MSc student in Engineering Responsible AI Systems at IMC Krems, where I also earned my BSc in Informatics.

See projectsGet in touch

Projects

Bachelor thesis, 2025-2026

Forecasting capacity consumption in manufacturing

Manufacturers plan capacity with static routing times that drift away from what actually happens on the shop floor. I built a pipeline that forecasts weekly capacity consumption per work center, trained on a three-year synthetic dataset (31,949 rows, 20 work centers), and compared five models from naive persistence to XGBoost.

0.355

Test R² for XGBoost with planning-state features from the ERP system

up from 0.204 without them, ahead of the ARIMA baseline at 0.271

R² on the log-transformed target, 460 work-center weeks in the test set.

Python, XGBoost, scikit-learn, pandas, NumPy

Read the thesis (PDF)

What drives the forecast

  • Planning state (forward-looking)
  • Historical (backward-looking)
What drives the forecast
FeatureTypeImportance
4-week average loadHistorical (backward-looking)0.188
Priority-weighted backlogPlanning state (forward-looking)0.107
Planned work (open orders)Planning state (forward-looking)0.070
Seasonal pattern (sin)Historical (backward-looking)0.065
Seasonal pattern (cos)Historical (backward-looking)0.046
Released but not startedPlanning state (forward-looking)0.037
Work in progressPlanning state (forward-looking)0.036
Last week’s loadHistorical (backward-looking)0.034
Load 4 weeks agoHistorical (backward-looking)0.032
Total plant loadHistorical (backward-looking)0.030
Due next weekPlanning state (forward-looking)0.019
Average efficiency (past)Historical (backward-looking)0.015
Downtime rate (past)Historical (backward-looking)0.015
Average scrap rate (past)Historical (backward-looking)0.013
4-week load volatilityHistorical (backward-looking)0.011
Idle week flagHistorical (backward-looking)0.005
XGBoost feature importance on the training set. The five planning-state features carry 37.2% of it. Source: thesis, figure 4.5.

Other projects

Local LLM with vector search for news retrieval

2025

A locally run LLM pipeline that uses embeddings and a vector database to index and retrieve product and news updates, with a small API on top for integration and testing.

TypeScript, FastAPI, Docker

Soil quality and sediment modeling (capstone)

2025

Cleaned and engineered tabular borehole data, then trained Random Forest and XGBoost models against a neural-network baseline for sediment classification.

Python, scikit-learn, XGBoost, PyTorch

Austrian flood monitoring system

2024

Real-time environmental monitoring platform for flood prediction and early warning. I led development as Scrum Master and worked on data collection and visualization.

React, Node.js, data visualization

Music streaming app

2024

A responsive streaming app built for a course on MVC, ORMs, and audio delivery.

Python, Django, SQLite

Movie data dashboard

2024

A personal dashboard for tracking movies and certification practice, built to get fluent with Dash and data wrangling.

Python Dash, pandas, APIs

About

I studied Informatics at IMC Krems with a focus on data science. My bachelor thesis grew out of my work as an AI/IT junior consultant at Concircle from February 2025 to June 2026, where I worked on SAP ERP digitalization projects.

Consulting showed me how machine learning behaves under real constraints: messy ERP data, business deadlines, and people who need to trust the output before they act on it. That is why I’m now doing an MSc in Engineering Responsible AI Systems. I want to build models that are useful, explainable, and worth trusting.

From
Romania
Based in
Lower Austria
Languages
Romanian (native), English (C1), German (B2)
Interests
AI ethics, industrial IoT, digitalization

Experience and education

  1. 2026-2028

    MSc, Engineering Responsible AI Systems

    IMC Krems University of Applied Sciences, Krems, Austria

    Master’s programme on responsible AI development, connecting machine learning systems, ethics, and engineering practice.

  2. Feb 2025 - Jun 2026

    AI/IT Junior Consultant (SAP ERP)

    Concircle Österreich GmbH, Vienna, Austria

    Digitalization projects in a consulting setting. I analysed processes, turned business requirements into IT solutions, and integrated machine learning for automated reporting and predictive outputs.

  3. 2023-2026

    BSc Informatics, data science focus

    IMC Krems University of Applied Sciences, Krems, Austria

    Thesis: Improving Production Planning using Machine Learning. Coursework in big data, AI, machine learning, databases, statistics, algorithms, software engineering, operating systems, and web development.

Skills

Programming

  • Python
  • SQL
  • R
  • C

Data science

  • scikit-learn
  • pandas
  • NumPy
  • XGBoost
  • PyTorch
  • Statistical analysis
  • MLOps
  • LLMs

Enterprise and infrastructure

  • SAP ERP (PP)
  • Docker
  • Kubernetes
  • Git
  • Linux

Web

  • React
  • Next.js
  • TypeScript
  • Node.js
  • FastAPI
  • Tailwind CSS

Contact

I’m open to data science roles, research collaborations, and conversations about responsible AI. Email is the fastest way to reach me.

[email protected]