Developer & Researcher

I work at the intersection of natural language processing, machine learning, and software development — building corpora and models on the research side, and shipping practical tools on the engineering side.

I hold an MSc in Cognitive Systems: Language, Learning and Reasoning from the University of Potsdam, with a strong foundation in Python and hands-on experience across NLP, data analysis, and web development. I bring these skills together with my interdisciplinary work in music and linguistics, developing applications that merge technology and the arts. Feel free to reach out through the contact section for inquiries.

Technical

• Languages
  Python · JavaScript · R · SQL · HTML/CSS · LaTeX · Git
• LLMs & generative AI
  On-premise deployment (Ollama, vLLM) · RAG pipelines · LangChain · FAISS · vector embeddings · prompt engineering (few-shot, chain-of-thought, function calling, structured output)
• Machine learning & NLP
  Hugging Face Transformers · PyTorch · TensorFlow · scikit-learn · spaCy · NLTK · BERT fine-tuning · sentence-transformers
• Multimodal
  Vision-language models (Qwen-VL) · speech-to-text (Voxtral) · document OCR
• Applications & infrastructure
  FastAPI · Flask · Gradio · Streamlit · Docker · Ubuntu · MySQL · n8n workflow automation
• Research methods
  Multi-platform collection · corpus construction · annotation protocol design with inter-annotator agreement · stance and sentiment classification · thematic and framing analysis · network and influence mapping · model evaluation

Experience

• Founder & Lead AI Architect
  Discursa · Berlin · 2026 – present

  Arabic discourse intelligence for foundations, NGOs, broadcasters, and research institutions. Dialect-segmented analysis across four Arabic communities, with diaspora audiences tracked separately rather than merged; narrative and framing analysis; stance distribution, volume timelines, influence mapping, entity detection. Multi-platform collection across Facebook, YouTube, TikTok, and the web. I retain technical ownership of every deliverable and engage specialist consultants — editorial, regional, and additional language capacity — on a project basis.

• AI Consultant
  Scitech IT Solutions GmbH · Sep 2025 – Mar 2026

  Production NLP systems: RAG pipelines, on-premise LLM infrastructure with Ollama and vLLM, document OCR with Qwen-VL, and speech-to-text with Voxtral — built for contexts where source material cannot leave controlled infrastructure.

• Researcher / Student Assistant
  University of Potsdam

  Corpus construction and NLP research, exploring where computational linguistics and machine learning meet.

Research & Publications

• Automated Analysis of Stance, Rhetorical Techniques, and Divisiveness in YouTube Comments on Political Party Videos
  M.Sc. thesis · University of Potsdam · 2026 · published open-source

  End-to-end pipeline for content analysis of polarised political discourse on a German-language corpus: corpus construction and conversation-thread reconstruction, annotation protocol design with inter-annotator agreement measurement, stance labelling at comment-pair and thread level, and transformer classifiers benchmarked against classical baselines with systematic evaluation and explainability.

• Syrian Rap: Political, Social, Gender, Artistic, and Linguistic Aspects
  Funded research report · 2022 · commissioned by Ettijahat, published by Mamdouh Adwan Publishing House
• Classical Music and Modern Syria: Education, Politics, War, and Migration
  2021 · cultural institutions in Syria before and after 2011, and the effects of conflict and displacement on practitioners
• BundesTube: A German Political YouTube Corpus with Zero-Shot Transcript Restoration
  Zaher Alkaei (University of Potsdam), Arthur Hilbert (Technical University of Berlin), Ronja Memminger (University of Potsdam) · in print

Selected Published Briefs

• Damascus Alcohol Restriction
  Discursa · March 2026

  880 comments from 534 unique voices on Syrian government platforms. Stance distribution, hour-by-hour volume mapping, and framing analysis establishing the discourse as a proxy contest over the character of the post-transition state. Levantine dialect, expert-translated verbatim.

• DW Jafar Talk: Channel Image Analysis
  Discursa · March 2026

  Audience reception of an international broadcaster’s Arabic output: 3,868 comments across 100 videos, with sentiment and dialect segmentation and seven recurring discourse themes identified.

Selected Projects

• PrivateGPT Zero — AI that runs locally on your device, right in the browser. No data sent to servers. (code)
• MIDI Graph — note-transition graph for MIDI and MusicXML. (code)
• PrivyBot — your local chatbot powered by Ollama and LangChain
• Sentiment Analysis of a sample of Taylor Swift’s songs (NLP)
• Argumentation Mining with BERT (NLP)
• Local Chatbot powered by Ollama
• The Arabic MIDI Keyboard II — browser-based keyboard and drum machine to play Arabic music
• Moving Musical Objects — generative music using moving visual objects

Writing on AI

• When Language Models Got Hands: Tool AI, Singularity Talk, and the Missing Animal Intelligence
• Arabic Music and the Challenges of Artificial Intelligence

Useful Resources

For machine learning, NLP, and web development:

• A cool visualization of a neural network in action