AngeloDe LeonSoftware
& applied AI
AI consulting & automation · San Francisco

AI that does real work for your business.

I build and deploy AI systems, connect them to company workflows, and train the people using them. My work spans on-site local LLM infrastructure, custom MCP servers, RAG applications, and company adoption of Microsoft Copilot, Anthropic Claude, OpenAI ChatGPT, and Gemini in Google Workspace. I work with individuals, small businesses, and enterprise teams from setup through handoff.

On-site local AI deployment
Custom MCP & RAG systems
Company rollout & training
San Francisco · remote collaboration
What I do

AI systems, from hardware to handoff.

Start with the workflow, data, access requirements, and people who will operate the system. Scope the infrastructure, integrations, training, and verification around those needs.

Local AI & on-site deployment

Plan and configure hardware for local LLM inference and AI services, from personal setups to SMB and enterprise environments. Set up model serving on the local network, harden the host and network access, and configure Tailscale or a comparable private-access solution for authorized remote users.

A local AI environment with defined access

Custom MCP servers & deployment

Build and configure Model Context Protocol servers that connect AI tools to approved company systems. Scope authentication and permissions, hosting, deployment security, monitoring, and maintenance so the integration has a clear operating owner.

Managed connections between AI and your systems

Custom AI applications & RAG

Build AI applications and retrieval-augmented generation systems around approved documents and business data. Connect ingestion, retrieval, model APIs, and application interfaces; define access boundaries, source references, evaluation tasks, and human handoffs around the use case.

Useful applications grounded in your information

Company AI rollout & training

Onboard employees and configure business or enterprise AI plans for small and large organizations. Implement and teach Microsoft Copilot, Anthropic Claude, OpenAI ChatGPT, and Gemini in Google Workspace, with role-specific sessions, account settings, data-handling guidance, and technical compliance-readiness work.

Configured tools and practical staff training

Claude Code & Codex for developers

Help engineering teams adopt Claude Code and OpenAI Codex in real repositories. Configure project instructions, approved MCP tools, permissions, and repeatable implementation, test, debugging, and review workflows. Turn a focused pilot into a documented team setup.

AI-assisted engineering workflows

AI-built project recovery & deployment

Take an AI-built website or application from an unfinished prototype toward a maintainable deployment. Review and clean up code, address security gaps, configure hosting and servers, and prepare an operational handoff. Industry-specific technical controls and authorized penetration testing are agreed in the project scope.

A clearer path from prototype to operation

Workflow automation

Connect email, forms, calendars, CRM, and business systems. Automate intake, follow-ups, document routing, and scheduling, with scoped access and human approval where actions need review.

Connected workflows with clear ownership

AI security & jailbreak assessments

Test authorized models and custom AI applications for jailbreak behavior, prompt-injection risks, unintended data exposure, and tool-permission failures within an agreed scope. Document reproducible findings, prioritize fixes, and retest the affected behavior.

Evidence-backed AI security findings

AI audits & governance readiness

Review AI tools, workflows, permissions, data handling, and human review practices. Turn findings into a prioritized remediation plan and operating guidance. Coordinate technical compliance readiness with the appropriate legal or compliance advisers; the engagement does not confer regulatory certification.

Documented controls and remediation priorities
In practice

AI for the people doing the work.

Explore a typical workflow. These are illustrative scenarios; the details and results depend on the agreed project.

Software & engineering teams

Before

Developers try AI coding assistants independently, with inconsistent setup, project context, and review standards.

A possible next step

The team pilots Claude Code and OpenAI Codex on scoped tasks, configures repository context and permissions, and uses its tests and code-review process before merging. The pilot becomes a shared setup and training guide, with results reviewed before wider adoption.

What it involves
  • Claude Code & OpenAI Codex
  • Repository setup & MCP
  • Testing & code review

Professional services

Before

Staff sort incoming email and retype client details into several systems.

A possible next step

An assistant drafts replies, routes documents, and prepares intake records for a person to review before anything is sent or committed.

What it involves
  • Approved document search
  • CRM integration
  • Human approval

Healthcare & wellness

Before

The front desk repeatedly answers questions about hours, services, and appointment preparation.

A possible next step

An assistant answers from approved public information and routes booking requests. Patient records stay outside the public assistant; sensitive workflows require a separate review.

What it involves
  • Approved FAQ content
  • Booking integration
  • Data boundaries

Retail & e-commerce

Before

Product descriptions, customer questions, and stock notifications compete for the same small team.

A possible next step

AI prepares product copy for review, retrieves answers from current policies, and routes low-stock notifications through existing systems.

What it involves
  • Shopify integrations
  • Content drafts
  • Workflow alerts

Real estate

Before

Website forms, listing inquiries, and email leads land in different places.

A possible next step

Inquiries enter one workflow, receive an approved acknowledgment, and create follow-up tasks with clear ownership.

What it involves
  • CRM automation
  • Lead routing
  • Follow-up scheduling
How it works

Small pilot first. Then scale what works.

01

AI readiness assessment & audit

Map your tools, data, and daily work. Review existing AI use, identify gaps, and deliver a prioritized plan with a focused use case, its risks, and the baseline it needs to improve.

02

Pilot project

Build one assistant, automation, or AI coding workflow. Test representative tasks and review the results against the existing process before expanding.

03

Rollout & training

Expand the useful parts, document permissions and settings, and teach the team how to operate the system.

04

Support & improve

Review reliability, costs, and new requirements. Ongoing support and changes are agreed as part of the scope.

Security first

Your data. Clear boundaries.

The model is only one part of an AI system. Access, retention, tool permissions, and review steps matter just as much.

  • Review vendor data terms and retention for the selected plan
  • Use scoped access, multi-factor authentication, and existing company logins
  • Keep sensitive information out of unapproved public tools
  • Document when a person must approve an action
ai-use-policy.yml · illustrative example
approved_tools: reviewed business plans
data: minimum necessary for the task
actions: scoped to approved systems
external_replies: human approval required
access: company login + MFA
review: accuracy, cost, and failure cases
Ways to start

Pick the step that fits.

Start with a focused review, a defined project, or ongoing support. The scope, costs, and responsibilities are agreed before work begins.

Step one

AI assessment

Fixed feeQuoted before work begins

Understand where AI can help and what implementation would involve.

  • Workflow and tool review
  • Data and access risk review
  • Prioritized plan and scope
Book an assessment
Keep improving

Ongoing AI support

Ongoing engagementSupport scope agreed together

Maintain the system and develop the next useful workflow as your needs change.

  • Reliability and cost reviews
  • Updates and access changes
  • Workflow improvements
  • Team enablement
Discuss ongoing support
Questions

AI, answered plainly.

Can you roll out AI for a small or large company?

Yes. I work with small businesses and larger organizations on platform selection, business or enterprise plan setup, employee onboarding, access settings, and role-specific training. Supported platforms include Microsoft Copilot, Anthropic Claude, OpenAI ChatGPT, and Gemini in Google Workspace. Controls and training follow the organization’s data and operating requirements.

Can you help our developers adopt Claude Code or OpenAI Codex?

Yes. I offer AI coding adoption consulting and hands-on training for engineering teams using Claude Code and OpenAI Codex. The work can include repository setup, project instructions, MCP integrations, tool permissions, testing, debugging, and code-review workflows. We agree on pilot tasks and evaluation criteria, then document a repeatable setup for your team.

What does an AI consultant help with?

I identify useful AI applications, evaluate tools and risks, build assistants and automations, and train teams. The work can start with an assessment or a specific implementation problem. You work directly with me from scoping through handoff.

What is included in an AI audit or readiness assessment?

We agree on the systems and workflows to review, then examine current AI use, data flows, permissions, vendor settings, output checks, and ownership. Deliverables can include an inventory, documented risks and gaps, prioritized recommendations, and a scoped implementation or training plan. It is a technical and operational review, not an assurance certification.

Can you help with AI compliance and governance?

I can help document approved tools, access controls, data-handling rules, human review steps, and evidence for your organization’s governance process. I coordinate technical readiness work with your legal or compliance advisers where needed. Applicable obligations depend on your business and use case; I do not provide legal advice or certify regulatory compliance.

Is our business data safe if we use AI?

That depends on the product, plan, settings, and data involved. I review vendor terms, retention, permissions, and integrations before connecting business information. For sensitive use cases, we evaluate private deployment options and decide what should remain outside the system.

Do we need to buy new software?

Often the first step is reviewing what you already have. Microsoft 365 and Google Workspace may offer relevant capabilities through your current plan or an add-on. I compare that path with custom development and document recurring costs before you commit.

Which AI tools and platforms do you work with?

My work includes Microsoft Copilot, Anthropic Claude, OpenAI ChatGPT, Gemini in Google Workspace, Claude Code, and OpenAI Codex. I also build custom MCP integrations, local model-serving environments, RAG applications, and automations with tools such as n8n and LangGraph. Platform choice follows the workload, data, access, and hosting requirements.

How long does an AI pilot take, and what does it cost?

The timeline and fee depend on the workflow, integrations, data readiness, and review requirements. An assessment produces a written scope, milestones, and cost assumptions. Pilots stay focused so we can evaluate a useful result before expanding.

Can you train our team or work with our existing IT provider?

Yes. I offer standalone workshops and training alongside implementation, and can coordinate access, security, and handoff with your internal team or IT provider. I am based in San Francisco and also work remotely.

Can you install and serve local AI on our network?

Yes. I handle hardware setup and configuration, local model serving, host hardening, and network access for personal, SMB, and enterprise requirements. Authorized remote access can use Tailscale or a comparable solution. Hardware, licensing, capacity, support responsibilities, and any external model connections are defined in the scope.

Can you build and host a custom MCP server?

Yes. The work can include server implementation, authentication, permission boundaries, hosting, deployment hardening, monitoring, and handoff. We define which tools and data the server may expose and who is responsible for operating it.

What does RAG implementation involve?

Retrieval-augmented generation connects a model to relevant approved information at request time. The work can include document ingestion, retrieval, access controls, source references, and representative answer-quality tests. It is distinct from training or fine-tuning the model’s weights.

Can you learn and qualify on the AI platform our company selects?

Yes. I have completed platform-specific training and certification for client adoption engagements. If your team selects a system, we agree on the vendor learning or qualification required, then scope implementation and employee training around it. Specific credentials and their current status are discussed for the selected platform.

Let’s make a plan

Your next step.
Made clearer.

Tell me what you are working on, what is getting in the way, and what a useful result would look like.

Based in San Francisco. Available for Bay Area projects and remote collaboration.