AI Researcher · System Architect · IT Systems Specialist

MSc student in Electrical Systems Engineering at Universität Paderborn — combining 14 years of enterprise IT architecture with active EC-funded AI research. Currently building sovereign AI infrastructure for European municipalities under the EU COLOURS Initiative, with a focus on responsible, GDPR-compliant, on-premise deployment.

📍 Paderborn, Germany  ·  ✉️ joys.learns@gmail.com  ·  LinkedIn  ·  GitHub

Internship thesis at Stadt Paderborn · EC-funded under the EU COLOURS Initiative

PadeRoBot is a sovereign AI reference framework for German municipalities — a production-grade, fully on-premise RAG system that answers citizen queries about city planning documents through a Telegram bot interface.

The project is part of the EU COLOURS Initiative (European Commission funded), which researches responsible AI deployment strategies across European public administrations. The goal: demonstrate that municipalities can run capable, compliant AI systems without any data leaving their infrastructure.

Architectural highlights:

• Hybrid retrieval — dense vector search (ChromaDB / HNSW indexing) + BM25 sparse retrieval, fused via reciprocal rank fusion
• Cross-encoder reranking — amberoad/bert-multilingual-passage-reranking-msmarco narrows candidates before LLM generation
• Local inference — quantised GGUF models via LM Studio / Ollama; zero external API dependencies
• Multilingual embeddings — intfloat/multilingual-e5-small for German-language municipal documents
• Production stack — FastAPI, streaming SSE, stateful chat sessions (SQLite), systemd service orchestration, Ubuntu 24.04 LTS, VMware virtualised server
• Compliance by design — EU AI Act risk classification, GDPR, BDSG, and OECD AI principles embedded into the architecture from the start

The hard constraint that shaped every decision: the model runs on-premise, sovereign, with zero external API calls.

• Sovereign AI architecture — designing LLM systems where data governance is a first-class constraint, not an afterthought
• RAG pipeline design — when hybrid retrieval beats pure vector search, why reranking is underrated, and what breaks in production
• EU AI Act in practice — translating risk classification and compliance requirements into actual system architecture
• Civic AI — what it genuinely takes to deploy AI responsibly in a public sector context
• Local vs cloud inference — privacy, latency, cost, and the governance implications of each choice

My background is broad by design — I have architected and operated systems across sectors, geographies, and scales.

AI & Applied Research (current) Building sovereign RAG pipelines and LLM infrastructure for public sector deployments under EC funding. NVIDIA Certified Associate in AI Infrastructure & Operations (2025).

Enterprise ERP & Digital Transformation Led Bangladesh’s first cloud-based SAP S/4HANA agile rollout (100M BDT programme) at Sun Gas Limited — covering ERP configuration, live production migration, user onboarding, and post-go-live stabilisation. Contributing factor to a measured 32% revenue increase (1.6B BDT / ~15M USD).

Cybersecurity & SOC Operations Managed full enterprise security stacks: TM Vision XDR (advanced threat detection), Fortinet Firewall, Cisco networking, SOC operations, disaster recovery and BCP planning.

Cloud & Multi-Cloud Architecture Designed and administered multi-cloud environments (Azure, AWS, VMware) for 200+ users across multiple geographies. Microsoft Certified Azure Administrator.

Infrastructure & Systems Administration 14 years running enterprise infrastructure — Windows Server, Linux (RHCE certified), Cisco (CCNA certified), MikroTik, Hyper-V, VMware, Proxmox — across BPO, energy, telecoms, and public sector environments.

Notable deployments: Police HQ network infrastructure · Bangladesh government projects · Embassy IT systems · 500-node ISO-18295 certified BPO environment · Terrestrial Media Monitoring System for Bangladesh Police

• NVIDIA Certified Associate — AI Infrastructure & Operations (2026)
• Microsoft Certified — Azure Administrator Associate
• Red Hat Certified Engineer (RHCE) & RHCSA
• Cisco Certified Network Associate (CCNA)
• SAP Cloud ERP Onboarding · IT Project Management (PM Aspire)

AI & Data Python · FastAPI · ChromaDB · LlamaIndex · LM Studio · Ollama · HuggingFace · GGUF/GPTQ quantisation · TensorFlow · PyTorch

Infrastructure & Cloud Linux (RHEL/CentOS/Ubuntu) · Windows Server · Azure · AWS · Oracle Cloud · VMware · Proxmox · Hyper-V · Docker

Security & Monitoring TM Vision XDR · Fortinet · Cisco · Grafana · Nagios · SolarWinds · Pulseway

ERP & Enterprise SAP S/4HANA · ERPNext / Frappe · Microsoft 365

Networking Cisco · MikroTik · PfSense · Asterisk PBX · Tailscale

English (fluent) · Bengali (native) · Hindi (intermediate) · German (intermediate, actively improving)

Photography (lifestyle & nature) · Trekking & mountaineering · Tech writing on AI and IT trends

Open to conversations about sovereign AI architecture, civic tech, and responsible AI deployment in the public sector.