Private AI Software Engineering
I build software and AI architectures for companies that want to use AI as a productivity lever without losing data governance, security, verification, and the ability to actually ship systems to production.
AI accelerates. Systems thinking governs.
AI makes it faster to generate code, prototypes, documentation, tests, and analysis. But that speed shifts the bottleneck: producing more is not enough. You have to decide what to build, how to verify it, where to run it, how to protect it, and how to ship it to production.
I don't sell generic AI expertise. I sell AI-augmented technical coordination.
Positioning
I don't replace vertical specialists and I don't sell “AI magic”. I step in when software development, infrastructure, security, data, automation, testing, and delivery have to work together.
Not an “AI generalist”, not a prompt engineer at the core, not a know-it-all. The value is turning a business goal into architecture, verified code, deployment, and maintenance.
What it means
I build software, automation, and AI systems for companies. What changes the work is not AI itself — it is the way AI combines with engineering discipline. Three ideas govern everything you read on the rest of this page.
Generating code, analysis, documentation, and tests today takes a fraction of the time it took a few years ago. What used to take a week now happens in a conversation. This is real acceleration: production of technical artifacts at a speed manual work cannot sustain.
Acceleration alone shifts the bottleneck but does not remove it. Software, data, infrastructure, security, testing, and production remain parts of one system and must be designed together. Technical direction is the discipline of holding these domains together while AI accelerates each separately.
Every AI output is a hypothesis until it is validated — structurally, in tests, in runtime, under load, against security constraints. Deterministic verification is the pipeline that separates probabilistic generation from release decisions and produces traceable evidence in place of trust.
The cognitive exoskeleton
Cognitive exoskeleton: AI is worn, not delegated to. It increases the designer's capacity for analysis, synthesis, and verification — while the designer remains responsible for direction, boundaries, and critical choices.
Applied research
Two technical documents published as PDFs. The hardware that holds twelve continuous hours of local inference, and the software that deterministically verifies AI-generated artifacts before release. Same method, two layers of the system.
Paper · Hardware
Five hardware iterations. From the first failed test (180 seconds, 90°C, shutdown) to the revision validated over 12 continuous hours. Eight GPUs at 100%, thermal range 35–52°C, zero drift. CAD design, custom aluminum, 3D-printed components.
Paper · Software
Pipeline modeled as a deterministic function over the Cartesian product of artifacts and configurations. Five ordered stages, early-exit algorithm, threat model, cryptographic attestation. Validated on 100+ iterated releases, 2,446 tests across 9 blocks.
Further reading
Epistemology
The cost of producing software with AI goes to zero. The cost of knowing whether it is reliable does not. What changes, in practice, when the bottleneck moves from production to verification.
Real hardware
Fourth hardware iteration. Real workload, thermal profile, 32B model in production on affordable hardware. No PowerPoint: numbers.
Professional practice
How multi-layered technical direction becomes a codifiable practice, not just individual skill. The thread that runs across roles and contexts.
From generation to delivery
AI accelerates the production of code, text, analysis, tests and documentation. A generated output stays probabilistic until it enters a technical workflow: analysis, structure, constraints, verification, debug, review, validation. So I work on three classes of artifacts — software and automations, AI infrastructure and systems, complex technical documents — with the same principle.
AI accelerates. Method governs. Verification makes it shippable.
Services
Concrete services to bring AI into real systems, without losing technical control.
Design and development of custom software with controlled use of AI for analysis, prototyping, development, and testing.
Read moreControlled conversational interfaces for documents, procedures, manuals, contracts, tickets, and corporate knowledge bases.
Read moreArchitectures in which models, data, and applications sit in the right place: cloud, private, on-premise, or local.
Read moreTechnical gates that turn AI-generated or AI-accelerated artifacts into verifiable, shippable software.
Read moreDocker, Kubernetes, CI/CD, hardening, logging, monitoring, and deployment to take prototypes to production.
Read moreAssessment of goals, data, risks, architecture, feasibility, costs, and minimum delivery path.
Read moreMethod
A clear path: analysis, architecture, AI-assisted prototype, verification, hardening, production, and iteration.
Problem, actors, constraints, data, environment, risk, and desired outcome.
Pattern, stack, boundaries, cloud/local, security, and technical roadmap.
Rapid MVP to validate flows, assumptions, and operational interfaces.
Gates on structure, policy, tests, runtime, dependencies, and security.
Deployment, logging, monitoring, backups, credentials, and operations documentation.
Evolutionary backlog, feedback, new releases, and continuous improvement.
Analysis
With AI, technical analysis stops being a preliminary step and becomes a continuous asset. Correlations, syntheses, and explorations at a depth that manual work — even with a team — cannot sustain within real project timelines.
Not impossible by hand. Just no longer sustainable.
Knowledge Engineering
Technical documentation, white papers, patents, compliance dossiers. Artifacts of high cognitive complexity that turn distributed information into usable knowledge.
Workflow
The workflows, combined with AI, bring project governance and quality assurance to levels that wouldn't be achievable by hand without automation. Not the individual tool — the workflow.
AI-generated code is a hypothesis. It must be verified before it becomes software.
Where the model runs is a data governance choice.
I design AI architectures where models, data, and applications are placed in the right environment: public cloud, private cloud, on-premise, or local.
Credibility
Concise proof points, without turning the page into a full CV.
Experience across infrastructure, security, development, DevOps, and critical systems.
Architecture, containers, Kubernetes, automation, and verification.
RAG, local LLMs, GPUs, private cloud, and controlled environments.
Applied research, complex systems, and intellectual property.
We start with a technical assessment: goal, data, constraints, environment, risk, and the minimum path to a verifiable prototype or a system in production.
