Portrait of Alexandru-Mihai Toboșaru

Alexandru-Mihai Toboșaru

I like to build AI agents and train models on real data


2026
SAP

I was an AI Developer (student practicum, April to June) at SAP Romania in Bucharest. I built a Product Data Quality Agent that scans large product catalogs, detects quality issues such as missing descriptions, wrong categories and incomplete attributes, auto-applies high-confidence AI-generated fixes and routes medium-confidence ones to a human review queue.

  • An MCP server (Node.js, Express.js) exposes SAP Commerce OCC APIs as agent tools, so the agent reads product data and applies fixes through ImpEx imports on its own.
  • SAP GenAI Hub generates the content with function calling. The agent runs on SAP BTP (Cloud Foundry) and talks to people through SAP Joule.
  • For SAP InnoWeeks I built an HR Performance Agent on SAP CAP and SuccessFactors OData, so HR managers can ask about team performance, KPIs and attrition risk in plain language. It won the Persona Advisory Excellence Award.
2025
Adobe

At the Adobe Innovate4AI Hackathon our team of three built Split AI, a Photoshop UXP plugin that turns natural-language instructions into structured layer edits through Gemini, with human approval before anything is applied.

2025
Raspberry Pi

I launched studentxyz, a web platform where ASE students track their grades and get ML-based predictions for future grades and dropout risk. It is a Flask app served from a Raspberry Pi 5.

2024 - 2027
Bucharest University of Economic Studies

Bachelor in Economic Informatics at the Faculty of Economic Cybernetics, Statistics and Computer Science, ASE Bucharest.

2020 - 2024
Ion Neculce National College

Baccalaureate, Mathematics and Computer Science at “Ion Neculce” National College, Bucharest.

bio

Alexandru-Mihai Toboșaru is an AI developer and an Economic Informatics student at ASE Bucharest. At SAP Romania he built AI agents on SAP BTP for product data quality and HR analytics. Outside of work he builds things like a Photoshop plugin on Gemini, a grade prediction platform for students and credit risk models. He speaks Romanian (native) and English (full professional proficiency) and lives in Bucharest, Romania.

featured projects

certifications

skills

Programming
  • Python
  • JavaScript (Node.js)
  • SQL & PL/SQL
  • Java
  • C++
  • HTML/CSS
AI / ML
  • Scikit-learn
  • XGBoost
  • TensorFlow
  • Pandas
  • Generative AI
  • Prompt Engineering
SAP Ecosystem
  • SAP BTP
  • Commerce Cloud
  • GenAI Hub
  • Joule
  • SAP Build
  • SAP CAP
Tools
  • Git/GitHub
  • Linux
  • MCP Protocol
  • REST APIs
  • Jupyter Notebook
Languages
  • Romanian, native
  • English, full professional proficiency

pet projects

All of these live on my GitHub.

split-ai is an Adobe Photoshop UXP plugin that translates natural-language instructions into structured, deterministic edits via Google Gemini: resize, crop, blur, colour and sharpness, plus layer operations. Gemini is held to a strict JSON schema and nothing runs before the user approves the plan.

studentxyz is a platform for ASE students with predictions for academic performance and progress analysis, live at tobosaru.xyz/studentxyz. Flask, SQLite, plain HTML, CSS and JavaScript, hosted on a Raspberry Pi 5.

studentxyz-ml-predictor is a TensorFlow classifier that estimates the chance of a scholarship and of graduating from a student's grades, with a small desktop app that reads a CSV and writes the predictions to a spreadsheet.

credit-risk-prediction is an XGBoost model for credit default, trained on Lending Club borrower data combined with US Federal Reserve macroeconomic indicators (ROC-AUC 0.72). Also a Kaggle notebook.

job-web-scraper scrapes IT and economic job postings from Romanian job boards (hipo, bestjobs, ejobs, iajob, plus LinkedIn and Indeed) and exports them to JSON. Made for students at ASE Cibernetica.

GestiuneVanzari is a Windows Forms desktop app for store management: products, clients, categories and sales. C#, ADO.NET and SQL Server LocalDB with the repository pattern and a database that sets itself up on first run.