I help companies become AI-native — redesigning engineering workflows so AI is built in by design, with humans owning truth, risk, and direction — and I build the open-source tools that make it work.
By day I work on AI agentic systems and security infrastructure for enterprise platforms. I also deliver talks, workshops, and hands-on adoption engagements for teams starting the journey — with a security engineer's habits throughout.
lobe: consulting
AI-enhanced is people using AI on the side. AI-native means the workflow itself includes AI steps by design — remove the AI and it stops working — while humans keep ownership of truth, risk, and direction. I help engineering teams make that shift.
Talks & workshops
"Becoming an AI Native Company" — a beginner-friendly session that takes a whole team from "what is an LLM?" to a working AI-native operating model: context and memory, spec-driven development, cross-LLM review, harnesses, and skills. Delivered on-site or remote, adapted to your stack.
Team training
Teaching employees to actually use AI — hands-on sessions that take developers and the people around them from first prompts to daily AI-native habits: project context files, durable specs, agent memory, review discipline. The goal is a team that works AI-natively without me in the room.
Adoption engagements
Hands-on installation of the operating model in a real team: project instruction files, durable specs, a shared human-curated knowledge brain, adversarial review gates, and the security guardrails to run coding agents safely.
The Enhanced Forge Flow
Write the spec with one strong author model, then review it in fresh sessions with one to four other LLMs — same family or different families. Feed the best critiques back to the author LLM, and implement only what both you and the author model agree is right.
Receive RequirementsHuman
→
Create BranchHuman
→
Write Specresearch · interviewone LLM · ex. Claude OpusAuthor LLM + Human
AI coding tools lose their plans when the session ends. Specmint turns ephemeral plans into durable, resumable specs stored in your codebase — a 6-phase forge workflow with deep research and developer interviews, TDD-first variants that enforce red-green-refactor, and rich HTML spec documents. Works with Claude Code, Cursor, Windsurf, Cline, Codex, and Gemini CLI.
$ npx skills add ngvoicu/specmint-core -g
$ npx skills add ngvoicu/specmint-core-html -g
$ npx skills add ngvoicu/specmint-tdd -g
$ npx skills add ngvoicu/specmint-tdd-html -g
related: [kluris] — pairs with Kluris brains for research-phase team knowledge
AI agents start every session cold. Kluris gives them a brain: a git-backed repo of human-curated knowledge — lobes, neurons, synapses — shared across every project and agent on the team. Agents propose; nothing is written without explicit human approval. Ranked BM25 search, no embeddings, no external calls. Open source (MIT), on PyPI. This page is laid out like one of its neurons.
$ pipx install kluris
related: [specmint] — ships Specmint companions per brain
A second-opinion layer for AI coding sessions: keep a roster of named participants and consult another coding agent — Claude Code, Codex, OpenCode, Pi — without leaving your session. Each consult is an isolated one-shot subprocess with a session handoff, workspace-confined tools, and an explicit consent gate before any of its changes are kept.
$ claude plugin marketplace add ngvoicu/consensflow-cc
$ pi install https://github.com/ngvoicu/consensflow-pi
related: [claude-code, codex, opencode, pi] — one question, many participants, consensus built one answer at a time
lobe: work
Dell TechnologiesAI Native Change Champion
AI agentic systems, MCP-based tool interfaces, RAG architectures, and orchestration layers that let LLMs interact safely with enterprise platforms. Reusable security infrastructure across microservice ecosystems: authentication and authorization libraries, identity provider integrations, standardized security patterns for Kubernetes.
ZeeSpire Software Solutionsfounder / independent
Full-stack SaaS products built end-to-end: Atlas (enterprise RAG chatbot — multi-stage retrieval, Google ADK orchestration, Qdrant, Keycloak), Burn The Burnout (multi-tenant employee wellness platform — Spring Boot, dual-auth RBAC, PostgreSQL tenant isolation), and Worriless (cross-platform task management).
Dell TechnologiesMassMutualLuxoftASMLUBS
lobe: previously
a40s01eperf(dell): reduce data import time from 40 minutes to 40 seconds
