Michael Nelz

This project develops a scalable, probabilistic, and highly reliable cashflow forecasting system to support strategic financial planning. By leveraging Bayesian modeling with Pyro, automated MLOps processes, and model explanations powered by GenAI, it generates precise, transparent, and uncertainty-aware forecasts. The solution strengthens decision-making, risk assessment, and financial resilience across the entire organization.

Techstack: Python, Pyro, Prognose, Scrum, Azure Synapse, Azure Open-AI, LLM, Prompt-Engineering, Langchain, Agentic AI, Plotly-Dash, Docker, FastAPI, Git

Creation of an end-to-end workflow tailored to accurately predict demand patterns and behaviors. Two-Pronged Approach for Intermittent Demand Forecasting

1. Binary Prediction: Identifying the occurrence of demands.

2. Regression: Estimating the magnitude of the demand.

Leveraging Large Language Models (LLMs) and Azure OpenAI to implement sentiment analysis for visit reports, enhancing qualitative insights.

Creation of an AI-Agent System to Explain Demand-Behaviour to the Planners.

Techstack: Python, API, Git, Prediction, Scrum, Kanban, Azure Synapse, CI-CD Pipelines, Azure Open-AI, LLM, GPT-4o, Prompt-Engineering, Langchain, Docker

Development of a robust, scalable, and highly accurate car price estimation model using advanced data science techniques by leveraging the advanced capabilities of Google Cloud Platform (GCP) and VertexAI. Automization of time-consuming national-vehicle-code matching via fuzzy matching supported by LLMs (GPT-4o and Gemma via Huggingface).

Proof of Concept for car-equipment-matching via LLMs to standardize the equipment overview on the website.

Proof of Concept for a car-suggestion Chat-Bot to help customers inform themselves about cars that fit their needs and suggest them fitting cars.

Techstack: Python, API, Webcrawler, Git, Prediction, Scrum, Kanban, GCP SQL, GCP IAM, GCP VertexAI, LLM, GPT-4o, Prompt-Engineering, Langchain, Huggingface, Docker