01
Secure AI Architecture
Production-ready RAG pipelines with strong security, governance, and reliability controls. Threat-modeled by default.
$ whoami
Alban Maria
$ cat ./role.txt
Cybersecurity & DevSecOps Leader · Forward Deployed Engineer · Enterprise AI Infrastructure · Secure RAG & Agentic AI Systems
$ ./status --short
> Building secure, scalable AI systems that solve real operational and business problems.
## about
For nearly two decades, I've worked at the intersection of cybersecurity, risk management, DevSecOps, application security, and application development — helping enterprises strengthen security posture through governance, secure engineering practices, threat assessment, penetration testing, and cloud-native AppSec consulting.
Over the past two years, I have been extending that foundation into AI automation, local LLM systems, and production-grade generative AI engineering. My focus sits deliberately outside the AI hype cycle — anchored in architecting secure, scalable systems that deliver measurable operational and business outcomes.
My work sits at the convergence of AI Security, DevSecOps, LLMOps, and Enterprise AI Governance. As enterprise AI adoption accelerates, organisations increasingly require practitioners who can both engineer these systems and ensure they are secured, governed, monitored, and operationalised with the discipline that production environments demand.
## focus
AI is moving from experiment to enterprise infrastructure — and the last mile is where it either works or doesn't. I work embedded with customer teams to deploy, harden, and operationalise Local LLM systems, secure RAG, and agentic architectures in the environments they'll actually run in. This is where DevSecOps evolves into secure AI delivery pipelines — LLMOps, AI governance, risk management, and inference architectures that hold up in production, not just in demos.
02
Local LLM & Self-hosted AI
Privacy-focused environments. Fine-tune Llama, Mistral, Qwen for domain-specific accuracy. Optimized inference and self-hosted deployment patterns.
03
Agentic AI & Automation
Workflow automation and agent systems for real enterprise operations — bridging commercial APIs with OSS ecosystems.
## live_projects
Shipped systems running in production — from enterprise infrastructure to side experiments.
LIVE · SIDE PROJECT
radioapp360.com ↗
RadioApp360
Spin a 3D globe, tap any country, and you're streaming its live radio in real time. Built for the small joy of hearing what a stranger on the other side of the planet is listening to right now.
$ open --new-tab radioapp360.com
IN DAILY USE · LOCAL-FIRST AI
runs on 127.0.0.1
myhomie AI
A local-first knowledge hub for every software and AI project on my Mac. It discovers project folders, builds a Confluence-style wiki for each one, and its Librarian answers questions with citations back to the exact source files — without sending code to the cloud.
Source folders stay read-only, secrets are redacted before anything is indexed, hosted models are strictly opt-in, and every generated doc waits for approval. A built-in planner tracks lifecycle, priority, and weekly hours across the whole portfolio.
OPEN SOURCE · API SECURITY
github.com/Alban-M/gabriel-releases ↗
Gabriel
From real browser traffic to secure, repeatable API tests. Gabriel records what an application actually sends over the wire, promotes any captured request into a credential-stripped test that replays with the session still intact, and runs the same file in CI.
Every release ships with keyless Sigstore signing, a CycloneDX software bill of materials, SHA-256 verified installs, and zero telemetry — designed so test artefacts are safe to commit in regulated environments.
$ open --new-tab github.com/Alban-M/gabriel-releases
LIVE DEMO · CRISISOPS
hcl-appscan-ai-assist.vercel.app ↗
AppScan Security Agent
HCL OpenAI Hackathon submission — an AI-assisted AppScan support consultant that helps customers and support teams resolve product issues faster through AppScan documentation, HCL KB articles, secure log redaction, and guided technical triage across normal and crisis operations.
This solution demonstrates how enterprise AI can strengthen customer experience, scale support operations, and create a reusable model for other HCL Software customer-facing products.
$ open --new-tab hcl-appscan-ai-assist.vercel.app/chat
COMPLETE · GOV TECH
tntenders.gov.in ↗
TN Tenders MCP
An AI-callable MCP server for Tamil Nadu's government procurement portal — search live tenders, score corruption risk across 7 dimensions, and pull vendor intelligence, all through natural language.
$ open --new-tab tntenders.gov.in
PRODUCTION · ENTERPRISE
private deployment
Core AI Gateway
Production-grade LLM routing layer — retries, rate limits, circuit breakers, and provider resilience that keep enterprise AI workloads online when individual model providers don't.
PRODUCTION · OBSERVABILITY
private deployment
Core AI ModelWatch
Continuous behavioral monitoring for production AI models — catches drift, latency anomalies, cost spikes, and policy leaks the moment they appear, before they surface as incidents.
## recent_impact
Selected outcomes from latest deployments — measurable, ready for real-time regulated environments.
1/10×
Legal document analysis — local
Self-hosted, fine-tuned Mistral system that outperformed GPT-4 on contract review at one-tenth the inference cost.
−60%
Agentic SOC automation
CrewAI-based framework that reduced Tier-1 analyst alert fatigue by 60% while improving true-positive detection.
0 egress
Financial-services local LLM
Privacy-preserving on-prem environment that eliminated data-residency risk while holding GPT-4-level accuracy.
## skills
AI & LLM Engineering
- AI & LLM Engineering
- Local LLMs
- LLMOps
- AI Infrastructure
- AI Automation
- Agentic AI
- RAG Architecture
- Fine-tuning
- Inference Optimization
- LangChain
- CrewAI
- AutoGen
- Open WebUI
- Prompt Engineering
- Custom GPTs
- AI Governance
- AI Security
- Enterprise AI
Cybersecurity & DevSecOps
- Cyber Security
- DevSecOps
- Application Security
- Secure Code Review
- Vulnerability Assessment
- Penetration Testing
- Threat Modeling
- Risk Management
- SOC Automation
- Security Governance
- Cloud Security
- Secure Architecture
Infrastructure & Engineering
- Python
- Docker
- Linux
- API Integration
- Self-hosted AI
- AI Workflow Automation
- MCP Servers
- Cloud-native Security
- Automation Engineering
## currently_building
Active projects keeping sensitive data inside the perimeter — no API leakage, no third-party exposure, GPT-4-level accuracy at fractional cost.
Local LLM Lab
Self-hosted environment running fine-tuned Llama and Mistral models for enterprise document analysis. Zero data egress, full auditability, GPT-4-level accuracy at a fraction of the cost.
Agentic AI Frameworks
Multi-agent systems built on CrewAI and AutoGen that automate security workflows, compliance reporting, and infrastructure provisioning — with human-in-the-loop governance.
Production RAG Pipelines
Retrieval-augmented generation systems engineered for production: hybrid search, re-ranking, source attribution, and hallucination detection built in.
Inference Optimization Engine
Inference optimization strategies and intelligent caching that push local-LLM performance into commercial-API territory — typical 40–60% cost reduction.
$ echo "If you're working on similar challenges, let's compare notes." && ./contact --brainstorm
## contact
Open to brainstorming on secure AI architectures, local LLMs, RAG pipelines, AI governance, DevSecOps, and agentic systems. Not a sales channel — a thinking channel.