Welcome! I am a Mechanical Engineer and Python Developer specializing in bridging the gap between deep technical engineering, data science, and modern AI-driven automation.
This repository serves as a public dashboard for my private projects in the fields of AI Orchestration, Signal Processing, and Process Automation.
Note: This showcase app is built with Streamlit to present the projects interactively. The actual projects behind each section use broader tech stacks (PyQt6, Flask, Dash, etc.) and live in private repositories.
- Project:
AI_orchestrator - Problem: Managing multiple AI agents and long-running tasks across different providers (Claude, Gemini, Codex) without manual intervention.
- Solution: A robust Python-based orchestrator with a Markdown-based queue system, automated provider fallbacks, and a Telegram integration for remote control and approvals.
- Advanced Features:
- Usage Suggester: Proactively suggests tasks (Vault-Skills, Git-Changes, failed retries) via Telegram when provider limits (Claude) are about to reset and capacity is unused.
- Heartbeat Mechanism: Continuous monitoring of system health, disk space, and git status, with automated summaries and cleanup tasks.
- Policy Engine: Fine-grained security layer (AUTO / APPROVE / DENY) with interactive Telegram approvals for risky operations.
- Analytics Dashboard: Built-in web dashboard (Chart.js) for tracking task success rates, provider utilization, and capacity timelines.
- Key Tech: Python 3.11+, Telegram Bot API, Multi-Provider Routing (Claude/Gemini/Codex), YAML-based Policy Engine, TF-IDF Memory System.
- Impact: Fully autonomous execution of coding, review, and documentation tasks with human-in-the-loop safety layers.
- Project:
energy_community_analysis - Problem: Managing decentralized energy communities (EEG) with thousands of members requires precise tracking of 15-minute consumption/generation data and accurate forecasting to optimize local energy usage.
- Solution: A comprehensive energy management system for a regional energy community (2,200+ metering points). It features a robust ETL pipeline, quarterly data partitioning in SQLite, and an interactive Dash dashboard for real-time analytics.
- Advanced Features:
- PV Forecasting: Hybrid ML approach using XGBoost and Prophet to predict solar generation based on 140+ features (temporal, astronomical, and weather-specific like fog risk).
- Weather Integration: Real-time data from Open-Meteo and Solcast, combined with clear-sky models for high-accuracy predictions.
- Smart Analytics: Automated calculation of self-sufficiency (Eigendeckung), grid interaction, and community-wide energy balance.
- Key Tech: Python (uv), SQLite (WAL-mode), XGBoost, Prophet, pvlib, Dash/Plotly, Optuna.
- Impact: Optimized local energy consumption and improved grid stability for an entire region.
- Project:
Aufbereitung_Zeitrohdaten - Problem: Raw vibration data from machinery (NVH - Noise, Vibration, Harshness) is complex and unorganized. Manual transformation of time-series data into frequency-domain insights (Campbell diagrams) is time-consuming and prone to errors.
- Solution: A high-performance automated pipeline for cleaning and transforming raw 15-minute sensor data into engineering-grade charts. It handles multi-sensor synchronization, Hanning windowing, and high-resolution FFT (Fast Fourier Transform).
- Advanced Features:
- Automated Campbell Generation: Converts time-series + RPM data into interactive Campbell matrices (Magnitude & Phase).
- Motor Order Extraction: Automatically extracts the first 5 motor orders (0.5 steps) for structural resonance analysis.
- FEM Integration: Generates
.femfiles (RLOAD curves) for direct import into Hypermesh/OptiStruct simulation workflows. - Phase-Reference Matching: Calculates phase relationships relative to a specific reference sensor for mode shape identification.
- Key Tech: Python, Signal Processing (FFT), NumPy, Hanning Windowing, Matplotlib/Plotly, FEM Export (RLOAD).
- Impact: 90% reduction in manual data preparation time. Enables rapid structural health monitoring and precise identification of resonance issues in cranes and automotive components.
-
Project:
EN13001_Tool-Suite - Problem: Manual crane statics calculations according to EN 13001 are complex, time-consuming, and error-prone. Legacy tools (Mathcad) were outdated and lacked modern integration.
- Solution: A comprehensive Python-based suite for norm-compliant (DIN EN 13001) design of crane components. It features a modern PyQt6 GUI, automated PDF reporting, and high-precision calculation modules.
-
Advanced Features:
- Multi-Module Architecture: Specialized calculators for Bolts, Welds (V2.0 fully norm-compliant), Fatigue, Pins, Wheel-Rail contact, and Elastic Stability (Buckling/Piping).
- Hot-Spot Integration: Advanced fatigue analysis using the Hot-Spot method (IIW recommendations) with automated FEM node interpolation and extrapolation.
-
Norm-Compliance: Rigorous implementation of DIN EN 13001-3-1 and 13001-3-3, including complex factors (e.g., weld quality factors
$\alpha_w$ from Table 8). - Professional Output: Generates detailed, audit-ready PDF reports with all calculation steps and norm references.
- Key Tech: Python 3.11, PyQt6, NumPy, Matplotlib, ReportLab (PDF), PyInstaller (EXE).
- Impact: 70% faster verification cycles for crane components. Direct transition from "raw FEM results" to "norm-compliant documentation".
- Project:
Engineering-Sales-Intelligence - Problem: High manual effort in lead generation, qualification, and CRM management for specialized engineering consultancies (FEM/CAE). Standard tools are too generic for deep-tech keyword requirements (e.g., distinguishing "EN 13001" from generic "statics").
- Solution: A full-stack automation pipeline that collects leads from RSS, job portals (JobPortal-A), and Gmail alerts, scores them with engineering-specific logic, and manages them in a custom CRM.
- Advanced Features:
- Two-Stage Scoring: Initial fast-scoring on metadata followed by deep-scoring via URL Enrichment (fetching full job descriptions only for promising leads) to save bandwidth and avoid rate limits.
- Intelligent Parsing: Uses Local LLMs (Ollama/Qwen2.5) and BeautifulSoup4 to extract structured data (salary, requirements, contacts) from messy email digests and HTML.
- AI Chat Assistant: A GUI-integrated chatbot (multi-provider: Claude/Gemini/Ollama) with 22 specialized tools to query the database, move leads, update CRM interactions, and perform company research.
- CRM & Orchestration: Full lifecycle tracking from "Scout" (lead detection) to "Analyst" (company research) and "Author" (automated Gmail drafts with engineer-focused tone).
- Multi-User Sync: Coordinated access via a Write Reservation System for OneDrive or remote access via Tailscale.
- Key Tech: Python 3.12, SQLite (WAL mode), Ollama, Gmail API (OAuth 2.0), Flask, Sentence Transformers (Semantic Search).
- Impact: 80% reduction in lead-sourcing time. Transition from manual browsing to a "Human-in-the-loop" approval workflow where the engineer only reviews high-value, pre-qualified leads.
- Project:
Boutique-SocialMedia&Vault-Skills - Problem: High manual effort in creating consistent, high-quality social media content and SEO-optimized product pages for a specialized retail boutique across multiple platforms (Instagram, Facebook, Pinterest, TikTok, YouTube).
- Solution: A specialized AI skill system integrated into an Obsidian vault that automates the entire content lifecycle—from professional image analysis to cross-platform copy generation and SEO optimization.
- Advanced Features:
- Multi-Platform Content Engine: Generates tailored posts for 5+ platforms (Reels, TikToks, Pins) from a single set of product images, including voiceover scripts and text overlays.
- Automated Image Branding Pipeline: A Python-based processing engine that automatically applies brand-compliant overlays, logos, and color grading to raw product photos, ensuring 100% visual consistency across all channels.
- Expert Image Analysis: AI-driven analysis of products (e.g., designer furniture, lighting, materials) by an "AI Design Consultant" to ensure technically accurate and emotional descriptions.
- Regional SEO Optimizer: Specialized WordPress/Stackable block generator that crafts SEO-rich copy focused on regional keywords and appointment conversion for showroom visits.
- Canva Integration: Automated design suggestions and direct asset preparation for rapid visual content creation.
- Key Tech: Python, Obsidian (Vault-Skills), OpenAI/Anthropic Vision APIs, WordPress REST API, Canva API, ElevenLabs (Voiceover).
- Impact: 95% reduction in content creation time. Professional, consistent brand presence across all digital channels with a focus on local lead generation (showroom appointments).
- AI/ML: Autonomous Agents, Local LLMs (Ollama), OpenAI/Anthropic APIs.
- Engineering: Signal Processing, Vibration Analysis (NVH), Structural Dynamics.
- Software Quality:
uv(Package Management). - Showcase App: Streamlit multi-page app with
@st.cache_datafor heavy computations, shared brand constants (COLOR_*,PLOTLY_DARK_LAYOUT) inui_shared.py, and no external runtime dependencies in the UI layer.
My work is driven by three core principles that bridge the gap between classical engineering and modern data science:
- Domain-First Data Science: I don't just apply algorithms; I leverage my background as a Mechanical Engineer to understand the underlying physics (e.g., vibration modes, solar radiation patterns) before selecting a model.
- Autonomous Reliability: Systems should work for humans, not the other way around. My projects focus on self-healing pipelines, automated fallback mechanisms, and proactive monitoring.
- Actionable Transparency: Whether it's a Campbell diagram for a crane or an energy balance for a community, the goal is always to turn complex "black-box" data into clear, actionable insights for decision-makers.
I focus on efficient, maintainable code that delivers measurable business value.
- Status: Available for freelance projects in Python automation, AI integration, and technical data analysis.
- Access: These projects contain proprietary IP and are kept in private repositories. Code access for specific modules can be granted upon request after an initial consultation.