About
Andrei Dmitrov
I help companies take AI from pilots to production.
As a solution architect I design how LLMs, retrieval and agents plug into the systems a business already runs, and the security, governance and cost model that keeps them running afterwards.
For the last five years I have done this at Siemens Energy. I took an enterprise knowledge-graph platform from MVP to a system the business runs on, owning the architecture end to end and coordinating up to 30 people. Since April 2026 I architect AI integration in the company’s AI squad: agents that answer business questions where half the answer sits in structured columns and half in free text, built next to the data so the access controls a company already has keep applying.
In parallel I built my own multi-tenant AI assistant platform end to end: retrieval over per-tenant knowledge bases, a provider-agnostic LLM layer, five channels including voice, native operator apps and GDPR by construction. It runs live today.
Before architecture I spent eleven years in networking and test engineering: Linux network stack internals, 5G and SIP, and test frameworks covering thousands of cases. That discipline is exactly what LLM systems need now.
Why Runko Works
Runko is Finnish for a frame: the load-bearing skeleton of a building, the hull of a ship. Karelian log houses stand for centuries because the frame holds them. That is the idea behind the name: get the architecture right, and what is built on top keeps standing.
How I work
Evaluation, not vibes
An AI system is tested like software: scenarios, hard checks, a judge, red teaming, before and after every change. Eleven years in networking and test engineering taught me that coverage is what a release stands on.
See the project →Build next to the data
AI goes where the data and its access control already are, instead of copying sensitive records into a new store that needs its own governance.
See the project →Foundations first
AI is only as good as the data it can reach. Scattered, inconsistent data makes a confident assistant that answers from the wrong half.
See the project →The right size of tool
A €11-a-month chatbot service is the right architecture for one studio; a multi-tenant platform is right for many businesses. Picking between them is the job.
See the project →I write the code I design
Architecture decisions hold up better when the architect still ships. I build the systems I design, from a solo platform to production agents.
See the project →
Timeline
- Jul 2026 – present
Own platform
A multi-tenant AI assistant platform, built solo.
- Nov 2021 – present
Siemens Energy
Lead architect of an enterprise knowledge-graph platform; since 2026 AI integration in the AI squad.
- Jul 2018 – Oct 2021
SIGOS
Mobile network testing: 5G integration, a performance-critical message aggregator, distributed SIP calls.
- Nov 2011 – Jun 2018
OKTET Labs
Network software testing at scale; team lead from 2015.
- Jul 2009 – Nov 2011
Institute of High-Performance Computer and Network Technology
Embedded networking for SpaceWire: hub firmware, PTP in the Linux kernel.
Stack
- Languages
- Python, TypeScript / Node, C, SQL, Bash
- AI
- Claude API, OpenAI API, RAG and hybrid retrieval, embeddings, agents and tool use, prompt caching, evaluation, usage and cost control
- Data & cloud
- PostgreSQL, Snowflake, knowledge graphs, ETL/ELT · AWS (EC2, VPC, S3, EKS, Lambda, API Gateway, Cognito, IAM, RDS, CloudFront), Terraform / OpenTofu, Docker, Kubernetes, GitLab CI/CD, NestJS, React
- Systems
- Linux networking internals, TCP/UDP, namespaces and routing, 5G / LTE / SIP, embedded
Languages
- Englishprofessional
- GermanB1
- Russiannative
Where
Nuremberg, Germany. Relocating to Dubai in December 2026. From then UTC+4: full overlap with European business hours, mornings with the US East Coast.
Get in touch