Start from constraints.
Understand the workload, risk, users, operating model, cost envelope, and failure modes before selecting technology.
CLOUD · PLATFORMS · AI-ASSISTED DELIVERY
AI is writing infrastructure faster than anyone can review it.
Archforge designs the architecture and the controls that make AI-assisted delivery safe to run in production: cloud platforms, infrastructure automation, and the verification layers that keep humans in command of systems that now build themselves.
00 · POINT OF VIEW
Complex systems should become easier to operate, not harder to understand.
Code review, change advisory, architecture review, security gates: every one of them was designed around the rate at which a human produces change. That assumption is now void.
The apparatus rarely fails loudly. It gets routed around quietly: a pipeline producing more change than the review step can absorb, an exception that becomes the default path, a control that still exists on paper and no longer describes what happens.
The answer is not to slow the model down or to ban it. It is to move the control point: from reviewing the output to constraining and verifying the generator. Boundaries the system cannot cross, verification that runs at machine speed, and enough observability that a human can still say what the system did and why.
Sophistication is not the goal. Complexity should be earned.
- CONSTRAIN THE GENERATOR
- VERIFY AT MACHINE SPEED
- KEEP HUMANS IN COMMAND
01 · APPLIED AI & AGENTIC SYSTEMS
AI requires engineering boundaries.
Models are powerful components. Production systems still need boundaries, verification, observability, and accountability. The work is engineering AI into real systems with controlled tools, explicit limits, useful context, and human oversight. The goal is dependable automation rather than autonomous theatre.
-
01
Constrain the generator
Decide what the model may actually touch: tools, scopes, credentials, environments, and the blast radius of a bad decision.
-
02
Verify before it lands
Policy, tests, and automated checks that run at the speed the change arrives, not at the speed a reviewer can read.
-
03
Make the execution observable
What ran, what changed, what it cost, and why, all recoverable after the fact by someone who was not watching.
-
04
Keep the decision points explicit
Automate the repetition, but leave the consequential choices inspectable and owned by a person.
02 · ENGAGEMENTS
Where to start
PRIMARY ENGAGEMENT
Agentic Delivery Readiness Assessment
Interviews across delivery, platform, security, and architecture, plus a review of the pipelines, infrastructure as code, and review gates you run today.
The output is a clear picture of how much your delivery process is already handing to machines, and what has to be true before it hands over more.
WHAT YOU GET
- how much autonomy each delivery stream actually has today
- governance gap analysis against that reality
- target-state architecture for controls and verification
- a 90-day roadmap, sequenced by risk
Start a conversation
TWO WEEKS · FIXED SCOPE
Architecture review
A paid half-day on one specific decision or design, with a written position at the end of it.
Fractional architecture
Ongoing technical direction, typically a couple of days a week, for teams that need architectural judgment without a full-time hire.
03 · FOUNDER
Razvan Hurhui
I have spent sixteen years on the same problem from progressively different angles, starting in SharePoint administration, then DevOps, cloud architecture, CTO, and enterprise architecture.
Most of that was inside organisations where the stakes were real: around six years at Accenture, latterly as a Software Engineering Associate Manager, along with work at PwC, Hewlett-Packard, Cegeka, Star Storage, and Ericsson. I was CTO and Head of Infrastructure at Noema Research. Today I am Senior Enterprise Architect at AD/01 and the founder of Archforge, working from Bucharest.
I use TOGAF, the Microsoft Cloud Adoption Framework, and the Well-Architected Framework as working instruments rather than checklists. Archforge is deliberately small: you talk to the person doing the engineering.
RECOGNITION
A project I built won Microsoft DevOps Partner of the Year, 2020.
It is not an award you apply for. Microsoft reviews delivery across its own partner portfolio and nominates from it. What that recognises is not effort but judgment: the engineering decisions held up under someone else's review.
04 · FOUNDATIONS
The engineering this is built on
Governing AI-assisted delivery is not a documentation exercise. It requires being able to build the underlying systems.
Cloud architecture
Production systems with clear boundaries and sensible trade-offs, across AWS, Azure, and Google Cloud.
Platform engineering & DevOps
Environments, CI/CD, release controls, and the guardrails that make delivery repeatable.
Infrastructure as code
Infrastructure as reviewable, versioned systems rather than manual console state.
Cloud-native systems
Containers, APIs, event-driven and serverless architectures built to the platform's strengths.
Reliability, security & observability
Identity, least privilege, logs, metrics, traces, recovery, and cost visibility from the start.
Across the stack
TECHNOLOGY
CLOUD
01- AWS
- Microsoft Azure
- Google Cloud
INFRASTRUCTURE & DELIVERY
02- Terraform
- Bicep
- CloudFormation
- GitHub Actions
- Azure DevOps
- Jenkins
RUNTIME & COMPUTE
03- Docker
- Kubernetes
- ECS / Fargate
- Cloud Run
- Serverless functions
ENGINEERING
04- Python
- TypeScript / Node.js
- PowerShell
- Bash
- APIs
- Event-driven systems
DATA & INTEGRATION
05- PostgreSQL
- MySQL
- SQL Server
- Queues and messaging
- Object storage
- Managed cloud data services
AI
06- LLM integration
- Agent tooling
- Retrieval
- Evaluation
- Structured generation
- AI orchestration
- Controlled execution
ARCHFORGE R&D · IN DEVELOPMENT
Builder
Builder explores a different way to create and operate technical systems: start with the outcome, turn intent into an explicit technical specification, and move from plan to working infrastructure and software through a governed, observable engineering process.
The long-term goal is to reduce the distance between an idea and a production-ready system without removing the controls that make production engineering trustworthy. The verification gate is the point of the design, not a stage bolted onto the end of it.
- INTENT
- SPECIFICATION
- BUILD
- VERIFY
- OPERATE
- describe what should exist
- make requirements and assumptions visible
- produce an implementation-ready plan
- automate engineering work under defined constraints
- surface progress, cost, risk, and results
- keep technical depth available when it matters
Builder is not intended to be another chat interface that happens to generate code. It is being designed around the harder problem: how AI-assisted engineering can produce real systems while keeping execution understandable and controlled.
05 · PRINCIPLES
How Archforge thinks about engineering
Five working assumptions behind every architecture decision, automation choice, and AI integration.
Cloud is infrastructure, not magic.
Every abstraction eventually meets networking, identity, state, cost, and failure.
Automation should increase control.
Automate repeatable work, but keep the critical decisions explicit and inspectable. If automation makes a system impossible to understand, the problem has only moved.
The simplest architecture that satisfies the constraints usually wins.
Use sophisticated technology when the problem requires it, not because it exists. Complexity must justify its operational cost.
Build for operations, and for the next change.
Deployment is not the finish line. Systems must expose state, failure, cost, and ownership, and they must still be changeable tomorrow.
Build something that needs to work.
If the problem involves AI-assisted delivery, cloud infrastructure, platform architecture, or the controls that keep any of it accountable, Archforge is interested in the difficult part.
Start a conversation