Automate a process,
secure what already runs.
Two engagements. An AI agent built for one specific business process, tested and delivered with its source code. Or an audit of what your systems and AI deployments actually expose, scoped by the OWASP, MITRE and ANSSI frameworks.
Claude Certified Architect
Claude Code, MCP and Agent SDK skills certified by Anthropic. AI agent expertise available at Colombani.ai for your projects.
Verify the credentialAutomate, or secure what already runs
Custom AI agent for a business process
One specific business process handed to an agent, designed by an Anthropic-certified architect (CCA Foundations). Tests, benchmarks, source code and monitoring included.
Discover → SecureSecurity audit and penetration testing
What your systems and AI deployments actually expose: attack surface, configurations, vulnerabilities, penetration testing. Covered by Colombani.ai professional liability insurance (IT security and intrusion testing).
Discover →What these solutions actually do
An agent that prepares your files
It reads incoming documents, extracts what matters, drafts the first version and flags what is missing. The professional reviews and decides instead of re-keying. Delivered in 3 to 6 weeks, source code included.
An agent that handles your inbound requests
It reads emails and forms, qualifies the request, routes it to the right person and drafts the reply. Your team only handles what needs a decision. Prototype in 2 weeks.
A picture of what your AI can actually reach
What data the deployed assistants can read, which connectors are open, which access rights stayed active after someone left. Prioritized report and remediation plan in 1 to 2 weeks.
A penetration test on your exposed perimeter
Attack surface mapping, vulnerability detection and controlled exploitation, accelerated by AI agents with systematic human validation. Report and evidence delivered.
The method
Scoping
What problem, what constraints, what tools in place. Deliverable: scoping document.
Proposal
Technical choice, scope, timeline, budget. Deliverable: detailed proposal.
Build
2-week sprints, demo at each iteration. Deliverable: solution tested in your environment.
Follow-up
Fixes, iterations, handoff. Deliverable: documentation + 30-day support.
The tools used
Claude API / Agent SDK
Core AI engine. Reasoning, generation, analysis.
MCP (Model Context Protocol)
Connection to business tools: CRM, databases, internal APIs.
Local models (Ollama, Mistral)
When data must stay on your servers.
Python / TypeScript
Backend and automation. Every project is delivered with its source code.
Agents and audits tailored to your profession
Professional secrecy, HDS hosting, ethics and strict GDPR constraints are built into the agents and audit scopes. See the approach by regulated sector.
Frequently asked questions
How much does a custom AI solution cost? +
Every project is quoted on a bespoke basis after a free scoping call. Scope depends on complexity, integrations, and desired level of polish.
Who owns the code? +
You do. Source code, documentation, and access credentials are delivered at the end of the project. No lock-in, no recurring license on the work delivered.
Can we use a local AI for sensitive data? +
Yes. We deploy local models (Mistral, Qwen via Ollama) directly on your servers or workstations. No data leaves your infrastructure. See the sovereign architecture offering.
How long for a prototype? +
2 weeks for a working prototype tested in your environment. Full deployment takes 1 to 3 months depending on scope.
Pricing
Describe your project. Response within 48 hours with an estimate and proposed approach.
Direct conversation with the founder.