Data Engineer | DevOps Specialist | Agentic AI & Quant Developer

I architect mission-critical systems for global enterprises and contribute to the core infrastructure of the modern cloud. My work bridges the gap between BaFin-regulated financial environments and cutting-edge autonomous agentic workflows.

I contribute to the foundational tools that power the internet's infrastructure:

• Hetzner Cloud CSI Driver: Contributor to the primary K8s storage interface for Hetzner Cloud.
• Google Research / TimesFM: Contributor to Google’s SOTA Time Series Foundation Model for zero-shot forecasting.
• Bitnami Helm Charts: Contributor to the industry-standard library for Kubernetes application deployment.
• Pandas-Dev / Pandas: Contributor to the world's most critical data manipulation library.
• Microsoft / Winget-pkgs: Active contributor to the Windows Package Manager repository.

S²-MoE — Evolving Sparse Spiking Mixture-of-Experts · 📄 Paper (DOI: 10.5281/zenodo.20846758)

A from-scratch, brain-inspired language model that gets cheaper as it scales — Top-1 spiking-expert routing, a matrix-state Gated Linear Attention backbone, and an offline "sleep phase" that merges, prunes, and grows its own network topology.

• ⚡ ~1/21 → ~1/43 the active FFN compute of a size-matched dense model, and up to 7.1× less serving energy (H100) — the efficiency lead widens as you add experts.
• 🎯 Quality-competitive with its distilled dense teacher (−5.15% perplexity on FineWeb-Edu, −4.4% on TinyStories at Nₑ=32, 3 seeds; BLiMP-neutral).
• 🔬 47 logged, fully reproducible experiments — including six documented negative results that bound the design space. Honest scoping over hype.
• 🧰 Built solo in PyTorch (cu128/Blackwell): custom autograd grouped-GEMM kernel, surrogate-gradient LIF neurons, SET/RigL structural plasticity, peer self-distillation.

Published preprint (CC BY 4.0) · independent research, no lab affiliation.

• Atruvia (Banking IT): Engineered high-availability OpenShift environments for Germany’s largest banking network, ensuring strict BaFin compliance.
  • Key Achievement: Mitigated a critical HashiCorp Vault cross-datacenter failure within a strict 2-hour window, preventing a mandatory federal reporting event with zero data loss.
• Automotive (HPE/VW/Porsche/BMW): Led petabyte-scale data replication across 60+ countries and designed hybrid cloud migrations.
• Health & Bio (Buchinger Wilhelmi / NavicareNow): CTO and Architect roles focusing on transferring sensitive health data to the cloud and building ML pipelines for fasting outcome prediction.
• Quant (Jim Harris Corp): Developed real-time stock prediction systems using Kafka, RabbitMQ, and Tensorflow.

• Specialized in Data Warehousing and high-profile innovation projects, including a VR Motion Seat for Formula E touring all over Europe.

I build, launch, and scale production-ready systems that solve complex algorithmic, hardware, and operational problems:

• AgentBureau: An elite B2B AI agency platform focusing on deploying production-ready multi-agent workflows, custom LLM fine-tuning, and operational automation.
• DocumentChat: Enterprise-grade AI RAG assistant for high-accuracy semantic search across massive document sets.
• WiFiSenseBox: An innovative IoT hardware-software solution engineered for signal detection, environmental sensing, and smart infrastructure metrics.
• AI Investing Bots: Quant and Algorithmic trading frameworks running automated forecasting models and deep quantitative analysis.
• PsychDiary: AI-powered companion mobile app for automated cognitive tracking and emotional analytics.
• NavicareNow: AI-driven medical navigation platform for patient documentation and guidance, backed by the Now2 Next Accelerator.
• Tiledom: A focused project within the Crypto, Web3, and economic simulation ecosystem.

• OpenShrimp: Persistent, task-driven LangGraph research agent with a modular plugin system and anti-bot evasion heuristics.
• AgenticSeek: Fully local autonomous agent that thinks, browses, and codes (Open-source Manus AI alternative).
• KuberneteslocalGPT: Private, on-premise RAG implementation tailored for isolated Kubernetes infrastructure.

• Python Tradingbot Framework: K8s-native algorithmic framework featuring 150+ technical indicators and distributed hyperparameter optimization.
• Kubernetes Autonomous Trading Agent: LLM-powered Solana agent running market scraping and transaction execution via native K8s CronJobs.
• AI Hedge Fund: Multi-agent collaborative team for algorithmic market reasoning and portfolio risk management.

• Python-OpenObserve: High-performance connector for sub-second OpenObserve telemetry and logging pipelines.
• Rspamd-Iscan: Advanced IMAP mailbox spam-filtering chart leveraging Bayesian learning models.
• Picture Rotation Fixer: Content-aware image processing pipeline using face detection algorithms to auto-orient scanned photos.

• Languages: Python (Tensorflow, Keras, FastAPI), Go (K8s CSI Drivers), Dart (Flutter Mobile/Web), Rust (Solana contracts), Java, TypeScript, HCL (Terraform).
• Orchestration & DevOps: Kubernetes, OpenShift, Helm, ArgoCD, Tekton Pipelines.
• Data Layers: Trino / Starburst, Kafka, RabbitMQ, PostgreSQL, MinIO, OpenObserve.
• Cloud Infrastructure: AWS Certified Solutions Architect, GCP, Azure.

• Primary Platform: DataFortress.cloud
• Inquiries: info@datafortress.cloud
• https://www.linkedin.com/in/justin-guese/