Doctoral Researcher (PhD) | AI for Engineering Systems & Intelligent Security Systems Munich, Germany · IEEE Member

TU Darmstadt — PhD: AI for Engineering Systems & Intelligent Security Systems ZM-I Group — Lead AI Engineer (AI-driven platforms for engineering & industrial applications) M.Sc. Cybersecurity (HDBW Munich) · OSCP · CEH · 10+ CVE credits

Research Focus

• Agentic AI & Autonomous Systems — multi-agent orchestration, tool-using LLMs, planner/executor architectures
• AI Automation — end-to-end workflow automation, document → decision pipelines, RPA × LLM
• Intelligent Security Systems — ML for threat detection, LLM red-teaming, adversarial robustness
• Intelligent Document Processing — OCR, layout analysis, Vision-Language Models
• AI for Engineering & Infrastructure — technical documents, maps, plans, industrial data
• Privacy-aware & Trustworthy ML — GDPR-compliant pipelines, on-prem inference, PII redaction

🛡️ Mythos Research Edition v2.0 — outside-in replication of Anthropic's Mythos Preview / Project Glasswing

• 8-phase pipeline (was 7) — build-sandbox · adversarial self-challenge · cross-session false-positive memory
• Powered by Claude Opus 4.7 · ~$1 per targeted scan
• Vendor-ready disclosure templates: GHSA · HackerOne · CVE · Bugzilla
• Research scaffold — live exec validator & exploit development stay private
• ArXiv preprint in preparation

📄 German-OCR 3.1 + German-Text 3.1 — local DSGVO document AI

• More compact, lower VRAM, same strict-JSON quality. Apache 2.0, GGUF.
• ollama pull Keyvan/german-ocr-3.1
• ollama pull Keyvan/german-text-3.1
• 🤗 Hugging Face: Keyven · 💻 GitHub: Keyvanhardani/german-ocr · 🌐 Hosted/On-Prem: german-ocr.de
• Next: German-OCR Word Plugin — local via Ollama / vLLM / llama.cpp or hosted API

🔐 April 2026 — 4 CVEs assigned · 9 more in triage

• 3× CVE-2026-40492 / 40493 / 40494 — CRITICAL in sail image decoders
• 1× MEDIUM in PhpSpreadsheet

• German-OCR 3.1 — local vision-language OCR for German documents (+ Text model 3.1)
• Mythos Research Edition — autonomous AI vulnerability-research scaffold (research-only)
• German-Privacy-Shield — GDPR-compliant PII detection & redaction
• German-OCR Word Plugin — local DSGVO document AI inside MS Word (coming soon)
• German-TTS — German text-to-speech (WIP / public release soon)
• Deutschland-Assistent — AI assistant for German document & workflow automation (coming soon)

💻 Languages · Python · C++ · Go · TypeScript · Rust · SQL · Bash

🤖 Agentic AI & Multi-Agent Systems LangGraph · LangChain · AutoGen · CrewAI · DSPy · OpenAI Swarm · ReAct · Reflexion · Tree-of-Thoughts · Planner / Executor · Tool-use & function calling · Multi-agent orchestration · Agent memory (short / long-term) · Self-critique loops · Browser-using agents · Sandboxed code execution · Agentic RAG

🧠 LLM Apps · RAG · Evaluation Weaviate · FAISS · Pinecone · Qdrant · Chroma · Hybrid + semantic search · Re-ranking · GraphRAG · Knowledge graphs (Neo4j) · Structured outputs / JSON mode · Prompt caching · Guardrails AI · NeMo Guardrails · LangSmith · LangFuse · Phoenix · RAGAS

🎯 Models · Training · Inference PyTorch · TensorFlow · ONNX · Transformers · PEFT · LoRA · QLoRA · DPO · ORPO · RLHF · TRL · Axolotl · Unsloth · DeepSpeed · FSDP · Hugging Face Hub · Ollama · llama.cpp · vLLM · SGLang · TGI · TensorRT-LLM · GGUF · MLX

👁️ Vision & Multimodal Vision-Language Models (Qwen-VL · LLaVA · InternVL) · OCR · Layout analysis · Document AI · CLIP / SigLIP embeddings

🔄 Automation & Workflows n8n · Temporal · Airflow · Prefect · Celery · Event-driven pipelines · Webhooks · Playwright · MS Office / Word automation

🛠️ Infra & MLOps Docker · Kubernetes · FastAPI · CUDA · GPU orchestration · CI/CD · MLflow · Weights & Biases

🔐 Security & Red-Team LLM red-teaming · Prompt injection (direct & indirect) · Adversarial ML · Vulnerability research · CVE workflow · GHSA · HackerOne · Bugzilla · GDPR / DSGVO · Threat modeling · OSINT · SBOM

📧 hardani@hotmail.de 🔗 LinkedIn 🤗 HuggingFace 🦙 Ollama 🌐 keyvan.ai · german-ocr.de

_{Open to selected collaborations in agentic AI, applied automation, vision-language systems, and intelligent security.}