I am Kristofer Jussmann (Ker102), an agent security and reliability engineer based in Estonia. I make AI agents safe to run and test whether they actually work: agent harnesses and evaluation, scoped agent permissions, AI-assisted infrastructure security, and secure cloud platforms on Azure and AWS.
Projects · Open source · Case studies · Portfolio · Blog · LinkedIn · GitHub activity
Last updated: October 2026
| Project | What it is | Result | Links |
|---|---|---|---|
| Harneloop Active open-source alpha |
Framework for improving agent harnesses: agents inspect what they produced, trace where it went wrong, test a candidate change in isolation, and keep it only if the test improves. | Agent improvement becomes a test discipline instead of trial and error; first applied to the ViperMesh Blender agent. | Case study · Project site · Repository |
| ViperMesh Active private product |
3D workspace with a Blender agent that works through about 130 constrained, validated tools and an MCP layer instead of writing open-ended Blender Python. | Against the Anthropic x Blender MCP baseline on the same model: faster on 6 of 7 live tasks and about 90% fewer agent tokens. | Case study · Open-source connector |
| nullstate-cli Active open-source alpha |
Purple-team CLI that attacks Terraform-defined infrastructure in local cloud emulators, applies a deterministic fix, and replays the attack to prove the fix holds. | On AWS LocalStack the exploit worked before the fix and was blocked on replay. The verdict comes from the replay, not the model. | Repository · Case study |
| Private Secure AI Automation Platform Archived |
Private n8n platform and OpenClaw personal agent on Azure Container Apps, with private networking, managed identity, passwordless CI/CD, and remote access over Tailscale. | Built and ran it, traced an idle-cost failure to my own design, tore it down with full exports, and documented a cheaper rebuild. | Case study |
| OpenRoot Active non-profit |
Developer community I founded: a developer program, a newsletter, and an open-source GitHub organization where members build in public. | 30+ members in the GitHub organization, a first partnership, and hackathons and workshops planned with partners. | openroot.dev · GitHub organization |
More work: PromptTriage (prompt research and evaluation), n8n Automation Atlas (workflow dataset and retrieval, dataset on Hugging Face), Crosswind Console, StoryForge, and Examplayer. The full portfolio has the rest.
| Project | Contribution | Status |
|---|---|---|
| Cap | Moved the caption word utilities into a shared cap-project module so the copies that had drifted apart now agree (#3295); a follow-up makes the number of words per caption segment configurable (#3296). |
1 merged, 1 open |
| OpenWhispr | Fixed dropped keyboard taps between rows on mobile (#2388); a second PR adds automatic API-key fallback for dictation (#2461). | 1 merged, 1 open |
| Prowler | Fixed a 404 in the OCI audit-retention check by creating the audit client in the tenancy home region, with regression tests. | Merged |
| Excalidraw for Obsidian | Proposed a Focus Embed Timer script, built on focus-embed-timer, my own zero-dependency timer for Obsidian, Excalidraw, and Markdown embeds. | Open |
| Graphiti | Traced non-OpenAI endpoint failures to client routing (#1339); a second PR makes the MCP read tools accept scalar group IDs (#1354). | Superseded upstream, 1 open |
| Mozilla Firefox | Working on a hard graphics driver bug in Firefox. | In progress |
| Matplotlib | Mentored a second-year computer-science student through their first upstream contribution, a merged documentation fix. | Merged (mentor) |
Founder, OpenRoot — present. Non-profit developer program, community, and newsletter with 30+ members in its GitHub organization.
Founder and Lead AI Engineer, Kaelux — 2024 to present. Research lab and consultancy for AI systems, automation, and cloud engineering work with collaborators and business clients.
Estonian Defence Forces — 2024 to 2025. Completed mandatory military service while continuing technical project development.
Credentials: Microsoft Certified: Azure Administrator Associate (AZ-104), DataCamp AI Engineering for Developers Associate, Anaconda Machine Learning Foundations, Docker Foundations, GitHub Foundations (GH-900), Microsoft Applied Skills: Azure Administration, and Cambridge C1 Advanced English. Next: SC-500 and Terraform Associate.
- Agent security: scoped agent permissions, constrained tool surfaces, MCP, attack-and-replay testing, threat modelling, and keeping the model out of the security verdict.
- Agent reliability and evaluation: agent harnesses, artifact-aware evaluation, trace-backed diagnosis, LLM-as-judge checks, and benchmarks that keep their negative results.
- Cloud and infrastructure security: Azure, AWS, Google Cloud, Terraform, LocalStack, Docker, Kubernetes and K3s, GitHub Actions, OIDC, managed identity, private networking, and cost controls.
- AI infrastructure: vLLM, self-hosted models, retrieval-augmented generation, vector search, and Hugging Face datasets.
- Languages and tools: Python, TypeScript, Rust, SQL, Bash, PowerShell, React, Astro, FastAPI, Three.js, and the Blender Python API.
View my GitHub profile · Browse all repositories
Personal site · Blog · Case studies · Portfolio
LinkedIn: Kristofer Jussmann · Hugging Face: Ker102 · Dev.to: Ker102 · Hashnode: Ker102
For collaboration or engineering work: kristoferjussmann@gmail.com