I am Kristofer Jussmann (Ker102), a DevSecOps, cloud platform, and agentic AI engineer based in Estonia. I build secure AI systems, cloud infrastructure, agent evaluation tooling, retrieval-augmented generation systems, and production automation with an emphasis on measurable results and public engineering evidence.
Projects · Portfolio · Case studies · Canonical blog · LinkedIn · Technical expertise · GitHub activity · Open source
Last updated: July 2026
Kaelux is my AI engineering research lab, parent company, and B2B consultancy. It develops the wider project portfolio and works with organizations that need practical AI onboarding, workflow automation, AI/ML integration, cloud deployment, and technical consultancy.
Kaelux has delivered automation and AI/ML work for multiple customers. Its products and subsidiaries have supported users across several domains, while its public work and service campaigns have built substantial reach on platforms including Instagram.
Infrastructure and cloud partners: Microsoft and Google Cloud
Supported by: Notion and PostHog
Visit Kaelux.dev · View the Kaelux repository · Review project evidence
| Project | Focus and engineering method | Evidence | Links |
|---|---|---|---|
| ViperMesh Active private product |
Unified 3D developer workspace and studio. Combines a browser 3D engine, neural 3D tools, and a deterministic Blender agent harness with artifact-aware evaluation. The Blender add-on and persistent MCP connector are maintained separately as open source. | 6 of 7 benchmark speed wins; 2.534× mean speedup; 90.91% lower acting-agent token use in a comparable task. | Case study · Open-source Blender connector |
| PromptTriage Active |
RAG-powered prompt research and engineering platform using structured evaluations and model comparisons. The current product is supported by a documented historical PR-gated Azure container-delivery path. | 28K+ prompt corpus; 1,080 structured evaluations; +13.9% relative instruction-following result in a scoped comparison. | Repository · Case study |
| Harneloop Active alpha |
Framework for self-evolving agent harnesses using portable harness units, artifact-aware tests, trace-backed diagnosis, isolated candidates, and evidence-gated promotion. | Public framework and case study; applied to the ViperMesh Blender benchmark program. | Project site · Repository · Case study |
| n8n Automation Atlas Active public project |
Schema-aware workflow generation, validation, dataset publishing, and retrieval experiments for n8n automation. | 36,985 import-ready workflows; 131,648-row ML dataset; 36,166 indexed vectors. | Explorer · Repository · Dataset · Case study |
| nullstate-cli Active open-source alpha |
Local-first purple-team CLI for Terraform. Replays cloud attack paths in LocalStack, uses a private vLLM endpoint for red/blue reasoning, and applies deterministic remediation. | AWS exploit replay succeeded before remediation and was blocked afterward; 104.45 tok/s recorded for the final red-agent run. | Repository · Case study |
| Private Secure AI Automation Platform Archived; recovered evidence |
Stateful n8n and OpenClaw platform migrated from a Google Cloud VM to Azure Container Apps with Azure Files, PostgreSQL Flexible Server, ACR, private endpoints, private DNS, managed identity, and Tailscale access. | Recovered topology, migration sequence, and cost evidence; missing deployment artifacts and attribution limits are disclosed. | Project record |
| Crosswind Console Paused public experiment |
Multi-agent research console using hybrid RAG, MCP, persistent HTTP pooling, and specialist travel, career, and trend agents. | 120+ real-time tools; documented tool latency reduced from roughly 60 seconds to under 3 seconds. | Repository |
| StoryForge Completed collaboration |
Voice-driven storybook system built with an observable agent-to-tool pipeline, safety controls, image generation, and Cloud Run delivery. | Five-person hackathon team; 122 backend tests; 8 of 9 recorded end-to-end checks. | Repository |
| Tuur360 Active pilot business |
Estonia-wide 360-tour, drone-media, and Google Maps visibility service with a documented capture-to-delivery workflow and evidence-led pilot model. | Live service site and published Google Maps media; client outcome reporting remains pilot-gated. | Service site · Google Maps contributions |
The full portfolio also documents the ViperMesh Azure GPU inference pilot, Examplayer, business-system pilots, and other engineering work with clear evidence boundaries.
| Project | Contribution | Outcome |
|---|---|---|
| Prowler | Fixed an OCI audit-retention 404 by creating the audit client in the tenancy home region and adding regression coverage. | Direct contribution merged upstream. |
| Graphiti | Traced a provider-compatibility defect affecting Gemini, Ollama, and vLLM-compatible endpoints and proposed generic-client routing. | Proposal independently validated and later superseded by the upstream factory refactor. |
| Matplotlib | Mentored a second-year computer-science student through a missing Sphinx Gallery thumbnail fix. | Student-authored documentation contribution merged upstream. |
Founder and Lead AI Engineer, Kaelux — 2024 to present. Client automation, AI/ML integration, workflow engineering, deployment, and advisory work.
Estonian Defence Forces — 2024 to 2025. Completed mandatory military service while continuing technical project development.
Current credentials include Microsoft Certified: Azure Administrator Associate, DataCamp AI Engineering for Developers Associate, Anaconda Machine Learning Foundations, Docker Foundations, GitHub Foundations, Microsoft Applied Skills: Azure Administration, and Cambridge C1 Advanced English.
- Agentic AI and machine learning: agentic systems, LLM orchestration, retrieval-augmented generation (RAG), corrective RAG, prompt engineering, model evaluation, LLM-as-judge workflows, vector search, Hugging Face, vLLM, LangChain, and LangGraph.
- Cloud, DevSecOps, and infrastructure: Microsoft Azure, Azure Container Apps, Azure Machine Learning, Azure Container Registry, Google Cloud, AWS, Docker, Kubernetes and K3s, Terraform, LocalStack, GitHub Actions, CI/CD, managed identity, private networking, observability, and cloud cost analysis.
- Data and automation: n8n, PostgreSQL, Supabase, Pinecone, Qdrant, Redis, synthetic dataset generation, workflow orchestration, semantic retrieval, and schema validation.
- Application and 3D engineering: Python, TypeScript, JavaScript, SQL, Bash, PowerShell, React, Next.js, Astro, FastAPI, Svelte, Vue, Three.js, Electron, Blender Python API, and Model Context Protocol (MCP).
- Engineering methods: artifact-aware agent evaluation, evidence-gated promotion, deterministic remediation, benchmark-driven development, PR-gated container delivery, trace-backed diagnosis, secure-by-design infrastructure, and reproducible technical documentation.
View my GitHub profile · Browse all repositories
Personal site · Canonical blog · Case studies · Portfolio
LinkedIn: Kristofer Jussmann · Hugging Face: Ker102 · Dev.to: Ker102 · Hashnode: Ker102 · Google Maps contributions
For collaboration, AI integration, or engineering work: kristoferjussmann@gmail.com