Enterprise Engineering Projects

Real-Time Time-Series Analytics & Data Pipeline Framework

Designed and engineered the core backend microservice infrastructure to support real-time time-series anomaly detection and clustering across 5 major German utility and environmental organizations:

  • Stadtentwässerungsbetriebe Köln (StEB Köln)
  • Rheinisch-Westfälische Wasserwerksgesellschaft (RWW)
  • Landesamt für Natur, Umwelt und Verbraucherschutz NRW (LANUV)
  • Emschergenossenschaft und Lippeverband (EGLV)
  • Bitcontrol GmbH (Bitcontrol)

Key Engineering Contributions:

  • Scalable Microservices: Developed a decoupled, event-driven microservice architecture containerized with Docker to process high-density data streams.
  • Data Streaming & Processing: Implemented Kafka and Kafka Streams to build reliable, real-time data ingestion pipelines and dedicated project-specific import adapters.
  • Stream Variations: Configured pipelines to adaptively handle varying data granularities (1-minute to 15-minute frequencies) and system latencies across up to 1,100 concurrent time series.
  • Data Lifecycle Management: Built robust scheduling mechanisms executing distinct execution workflows based on client readiness, supporting real-time streaming, scheduled batch processing, and historical backfilling/reprocessing for up to 15 years of archive data.

Intelligent Conversational LLM Pipeline & Intent Routing Engine

Conversational Commerce & Customer Support Automation

Designed and developed an early-stage conversational AI pipeline for a leading food and beverage delivery platform, enabling automated customer ordering, intelligent semantic search, and automated support routing.

  • Client / Platform: Snappfood (Food & Beverage Delivery Marketplace)

Key Engineering Contributions:

  • Intent Classification & Parameter Extraction: Built NLP pipelines to process raw customer chat text, classifying requests into operational intents (e.g., “food search”) while dynamically extracting search parameters and filtering criteria.
  • Context-Aware Recommendations: Integrated user profile data with intent parameters to query the product database, dynamically serving personalized restaurant and food recommendations directly within the chat interface.
  • Intelligent Support Routing: Developed an automated detection mechanism to identify customer complaints and seamlessly hand off complex order issues to live support team members.
  • Domain Fine-Tuning: Fine-tuned open-source transformer models using tailored, domain-specific instruction datasets to maximize classification and parameter-extraction accuracy under production constraints.

Solo Projects & Open-Source Contributions

  • A RAG-based Document Q&A Agent on Azure Developed an AI agent for document question answering, enabling users to interact with documents through natural language queries and retrieve relevant information using Azure AI Foundry and Document Intelligence.
  • Intelligent Receipt & Invoice Processing Android App on Azure Developed an Android app with a Microsoft Azure backend for automated receipt and invoice processing, extracting, categorizing, and structuring financial data from images and PDFs for reporting and expense management.
  • GitLab Management MCP Agent Built an MCP (Model Context Protocol) server that lets AI assistants like Cursor to create, update, label, and assign GitLab issues.
  • Digital Onboarding & Presence Mobile App for Small and Medium Businesses Built a mobile application designed to simplify and automate the digital footprint, online presence, and local marketing workflows for freelancers and SMEs.
  • Local-RAG: Retrieval-Augmented Document System Built a Retrieval-Augmented Generation (RAG) application for querying local documents using LangChain, a Chroma vector database, and ChatGPT.
  • FastAPI Azure Full-Stack Template Engineered a production-ready FastAPI backend and frontend boilerplate template optimized for containerized cloud environments (Azure/Docker).
  • BentoML: Serve & Deploy Machine Learning Models Streamlined deployment pipelines by packaging machine learning weights into containerized REST APIs and reproducible Docker images using BentoML.
  • Text2Image-NN Classifier: Fractal Analysis of Long Texts Developed a cross-domain NLP framework mapping long texts into visual structures to leverage pre-trained computer vision neural networks for classification.
  • Sequence Labeling System (fastai) Implemented an optimized sequence labeling pipeline over language models to process entire, long-context documents without token-length limits or memory constraints.
  • MorphoBERT: State-of-the-Art Named Entity Recognition (NER) System Awarded 1st place in the NLP Solutions for Under-Resourced Languages shared-task competition for designing a custom BERT and morphological pipeline.
  • Persianp: High-Performance Text Processing Toolbox Authored an enterprise-grade Java text processing library implementing optimized pipelines for core tokenization, lemmatization, and POS tagging.