Artificial Intelligence & Data Science Undergraduate @ UBI | AI Systems Engineering, LLMOps, Backend & Platform Engineering
I am an Artificial Intelligence and Data Science undergraduate at the University of Beira Interior (UBI), building independent portfolio projects around reliable AI systems, local-first runtimes, backend services, and evaluation workflows.
My work focuses on building reliable AI systems with deterministic execution, persistence, observability, operational tooling, and clear feedback loops for continuous improvement. I care about engineering discipline: systems should be understandable, testable, and useful beyond a notebook.
The projects below are self-directed portfolio work. They reflect my interest in turning AI ideas into practical software that can be inspected, operated, and evaluated.
A governed, reproducible visual-inspection runtime that verifies a PaDiM anomaly signal survives training, ONNX export, and ONNX Runtime. On the VisA proxy dataset, runtime equivalence across 6,492 samples recorded a maximum absolute deviation of 7.105e-15 against a 1e-12 tolerance—evidence of signal preservation through the integrated runtime, not accuracy, calibration, production readiness, or cross-domain validity.
Focus: ML Systems Engineering · Reproducibility · Runtime Verification · ONNX Runtime · Deterministic Replay · Provenance Evidence
Status: Completed Portfolio Project
Repository: github.com/andrejr03/kalibra
Live portfolio: andrejr03.github.io/kalibra
A local-first LLMOps runtime core focused on deterministic execution, persistence, evaluation, observability, prompt lifecycle, readiness, and operational tooling.
Focus: Local-First AI Runtime · LLMOps · Deterministic Execution · Persistence · Evaluation · Observability · Operational Tooling
Status: Completed Portfolio Project
Repository: github.com/andrejr03/prisma-llmops
Prisma v1 is a local-first LLM engineering platform focused on Retrieval-Augmented Generation (RAG), AI evaluation, prompt regression, runtime observability, and source-grounded AI workflows. It demonstrates practical engineering around retrieval quality, evaluation, and reliable LLM application development.
Focus: Retrieval-Augmented Generation · AI Evaluation · Prompt Regression · Runtime Observability · Source-Grounded AI
Status: Completed Portfolio Project
Repository: github.com/andrejr03/prisma
An AI system that reconciles invoices, purchase orders, and receipts, detects field-level discrepancies, preserves evidence, measures reliability, and routes uncertain cases to human review.
Focus: Document Intelligence · Reconciliation Systems · Reliability Scoring · Human Review · Explainable AI
Status: Completed Portfolio Project
Repository: github.com/andrejr03/crosswise
An AI evaluation and reliability project for source-grounded question answering over the EU AI Act. LexiBench combines a versioned benchmark, evidence-completeness scoring, retrieval evaluation, failure taxonomy, and a static evaluation workbench to measure how reliably AI systems answer source-grounded questions.
Focus: AI Evaluation · Reliability Engineering · Evidence Completeness · Grounded Retrieval · Responsible AI
Status: Completed Portfolio Project
Repository: github.com/andrejr03/lexibench-ai
A recommendation system that helps students find housing that fits their needs, combining a transparent fit-scoring approach with geospatial analysis. It frames housing search as a decision-intelligence problem — ranking options against personal priorities rather than a single price filter.
Focus: Recommendation Systems · Decision Intelligence · Geospatial Analytics · Student Fit Scoring
Status: Completed Portfolio Project
Repository: github.com/andrejr03/studentfit-bavaria
A football match prediction tool that pairs predictive modelling with clear, human-readable explanations for each forecast. The goal is not just to predict outcomes, but to surface why the model leans a certain way and how confident it is.
Focus: Machine Learning · Football Prediction · Explainable AI · Model Evaluation · Predictive Analytics
Status: Completed Portfolio Project
Repository: github.com/andrejr03/statsport
- LLMOps
- AI Systems Engineering
- Backend Engineering
- Platform Engineering
- AI Evaluation
- Machine Learning
- Recommendation Systems
- Explainable AI
- Predictive Analytics
- Geospatial Analytics
- Runtime & ML Systems: Python · ONNX · ONNX Runtime · NumPy
- Backend & Persistence: FastAPI · SQLAlchemy 2.0 · Alembic · Pydantic
- Quality & Delivery: pytest · mypy · Ruff · Git · GitHub · GitHub Actions
- Studying Artificial Intelligence and Data Science at UBI
- Building AI systems engineering and LLMOps portfolio projects
- Strengthening backend and platform engineering skills
- Deepening work on reproducibility, evaluation, runtime verification, and observability