I'm Liridonë Zhugolli Kamberi — a data scientist and machine learning engineer based in Berlin. I build ML that runs as real systems, not notebooks: models served in containers, deployed to the cloud, and wrapped in the orchestration and guardrails that make them genuinely useful.
My work spans three areas that increasingly overlap: modeling (deep learning, computer vision, audio, and multimodal systems), agentic AI (multi-agent orchestration, Google's Agent Development Kit, and MCP tool layers), and productionization (containerized serving, cloud deployment, and human-in-the-loop safeguards).
A few things I've built recently:
- Cold-Chain Inventory Optimization Agent — a multi-agent system that monitors perishable inventory and reasons over competing objectives (spoilage vs. stockout vs. holding cost) to produce a ranked list of reorder actions. An orchestrator coordinates three specialist agents on Google's ADK, backed by a dedicated MCP server, with a cost-threshold approval gate: cheap, reversible actions auto-execute, while expensive irreversible orders are escalated for human sign-off.
- Zero Bias — the core ML for a multimodal confidence-detection system for HR interviews, built during my data science internship at Zummit Infolabs. I trained complementary computer-vision and audio models and fused their outputs into a single, more robust confidence score (~92–95% accuracy).
- Scan Your Skin — a melanoma-detection web app I pitched and led a team of four to ship in ten days, combining a custom CNN with a VGG16 transfer-learning model, containerized with Docker and deployed on Google Cloud Platform.
Before data science, I spent years as a 3D digital artist, running an independent design practice and studying game art and sculpture. That background still shapes how I work: strong visual communication, attention to detail, and a design-driven approach to hard problems.
Skills
Languages: Python, SQL
ML / DL: scikit-learn, TensorFlow, Keras, computer vision, audio/speech, multimodal modeling
Data: Pandas, NumPy, SciPy, Statsmodels, BigQuery, Matplotlib, Seaborn
Agentic / LLM: multi-agent systems, Google ADK, MCP (FastMCP), LiteLLM, tool-use orchestration, human-in-the-loop gates
Deployment: Docker, Streamlit, Google Cloud Platform (GCP), MLflow, Git/GitHub
Education
- Data Science & AI — Intensive Bootcamp · Le Wagon, Berlin (2023–2024)
- Mathematics · University of Pristina — Natural & Mathematical Sciences (2008–2009)
- BA, Sculpture · University of Pristina — Faculty of Fine Arts (2008–2012)
- Game Art & 3D Animation · SAE Institute, Frankfurt (2015–2016)
Languages
English (fluent) · German (B1–B2, working proficiency) · Albanian (native)
Get in touch
Based in Berlin, Germany.
Email: [email protected]
GitHub: github.com/liridonezhk · LinkedIn: linkedin.com/in/liridone-zhugolli-kamberi