My role · Author & maintainer
Surreal-Memory: persistent memory for AI agents
Open-source persistent memory for AI agents, authored and maintained by Toni Nowak using Python, SurrealDB and MCP.
Read moreHello, I’m
Toni Nowak
Engineer by trade.
Curious by nature.
I build AI agents, untangle complex systems, and write about what I learn along the way.
AI Agent Engineer / Systems Architect / DevSecOps Consultant
Based in Norway · Working internationally
From my notebook
What I’m thinking about
AI / Agents
I Replaced a Process, Not a Person
How I automated 15 hours of weekly vulnerability triage while keeping security judgment, approvals and accountability with people.
· Toni Nowak
Read articleAI / Agents
Your AI Agent Just Got Hacked. And You Wouldn't Know.
A prompt-injection case study showing how malicious repository instructions can manipulate coding agents, plus practical defensive controls.
· Toni Nowak
Read articleProjects
A few things I’ve built
My role · System architect
Tender-Flow: multi-agent public procurement
A deployed multi-agent platform designed and built by Toni Nowak for structured Norwegian public procurement workflows.
Read more
About me
The person behind the work
I design and troubleshoot AI agents, backend integrations and the infrastructure they run on. I am the author and maintainer of Surreal-Memory and the architect behind Tender-Flow, a multi-agent platform for Norwegian public procurement.
15+ years across production IT, platform engineering and DevOps, including Sopra Steria, Sportradar and C&F. I build around clear scope, enforced access boundaries and independent verification.
Based in Trøndelag, Norway. Founder and AI/ML/Systems Architect at AI-Flow since April 2026; previously Senior Consultant at Sopra Steria (October 2024–May 2026). Available for implementation, diagnostics and technical advisory.
Experience across: C&F · Sopra Steria · Pfizer · UniCredit · LOT Polish Airlines
Experience
Career Journey
A track record of building resilient systems at scale across enterprise environments, from aviation IT to sports data platforms and consulting.
2024 – 2026
Senior Consultant at Sopra Steria
Previous role
- Delivered enterprise cloud migrations
- Architected AI-augmented workflows
2021 - 2024
Platform Engineer at Sportradar
Previous Role
2020 - 2022
Dev/Ops Engineer at C&F
Past Experience
- Automated CI/CD pipelines
Show full experience
2019 - 2020
Dev/Ops Engineer at Unilink S.A.
Past Experience
2017 - 2019
IT Systems Administrator at LOT Polish Airlines
Past Experience
What I work with
What I work with
AI Agent Engineering & Troubleshooting
Design, implementation and troubleshooting of AI agents, MCP, RAG, persistent memory and backend integrations with verifiable acceptance criteria.
Read moreAI Security & Architecture Review
A practical AI security and architecture review of trust boundaries, tool permissions, data flows and verification before agent autonomy expands.
Read morePlatform Engineering & DevOps
Platform engineering for Linux, Kubernetes, CI/CD, infrastructure as code and reliable backend services with reproducible operations.
Read moreExplore my full stack
AI/ML
- Claude
- LLMs
- RAG
- Agent Orchestration
- Prompt Engineering
Cloud
- AWS
- OpenStack
- Kubernetes
- Docker
- VMware
DevOps
- Terraform
- Ansible
- Puppet
- CI/CD
- GitOps
Security
- DevSecOps
- Vulnerability Management
- Compliance Automation
Languages
- TypeScript
- React
- Python
- Bash
- PowerShell
- SQL
Learning
- Rust
Certifications & Professional Training
Courses, credentials and continued learning
Anthropic Education · VMware · NetApp · AWS · Crunchy Data · MongoDB · IBM · SAP · Oracle · Euler · Marktechpost
Verified certificates and professional training in AI agents, MCP, cloud, databases and infrastructure. Exact names are preserved; course completion is not presented as an exam-based certification.
Explore all 21 credentialsThe conversation continues
Selected and reviewed 9 September 2026. Original dates and full text on LinkedIn.
Model access is an infrastructure dependency
Why access rules and vendor eligibility belong in the architecture review, alongside cost and privacy.
LinkedInSize inference from memory requirements, not headlines
Model parameter totals do not tell the whole story of residency, offloading and hardware sizing.
LinkedInA quarterly audit for AI subscriptions
Measure ordinary work, check the plan, test the fallback and compare with the previous quarter.
LinkedInShow all 10 posts
The time barrier for one-person businesses
A narrow, useful implementation can be more valuable than another general introduction to AI.
LinkedInVerification must produce a record
Independent checks, reproducible results and visible failures around agent outputs.
LinkedInEvaluate the harness as well as the model
Planning, retries and recovery can change outcomes; vendor benchmarks still need independent verification.
LinkedInMake quantization trade-offs measurable
The quality cost belongs next to the memory saving when choosing a production model.
LinkedInKeep the fallback tested and current
A backup model needs periodic evaluation, not just a permanent place in a configuration file.
LinkedInAgent identities need owners and a lifecycle
Inventory, credentials, expiry and revocation are part of deploying agents responsibly.
LinkedInEnforce tool permissions at dispatch
A list of tools in a prompt is not a substitute for authorization in the execution layer.
LinkedInSay hello
Let’s exchange ideas.
Have a question, an interesting challenge, or a different perspective? I’d like to hear it.
toni@nowak.sh