AI that answers from your real data and only does what you allow. I build it into the app and tools you already run.

Companies that need an AI engineer, CTOs and product teams with an existing app, CEOs who want AI in their product, and agencies with a client project.

Your customers or staff need answers from your own data (orders, products, documents), not from what a model guesses.

You want an AI assistant or agent that does real tasks, but only the actions you allow, with a person's approval where it matters.

Manual work such as order lookups, support triage or data entry takes hours that automation could save.

You cannot tell whether the AI gives correct answers, or whether a change made it worse.

AI assistants that answer from your orders, products, documents and database, with sources shown.

AI agents that use your app's tools, each tool with its own checks and only the permissions it needs.

MCP servers, so AI tools such as Claude and ChatGPT get safe access to your app.

RAG: search by meaning instead of exact words, with pgvector in your PostgreSQL database.

Workflow automation in n8n or Rails, connected through APIs and webhooks.

Evals and tracing: test runs with expected answers, and a record of every step the agent took.

AI Automation, from $600, delivered in about 1 week: one task done by AI, such as sorting tickets, drafting replies or reading documents.

AI Assistant, from $1,800, delivered in about 2 weeks: a chatbot that answers from your data with sources, or an MCP server for your app.

AI Agent, from $3,750, delivered in about 3 weeks: takes actions in your tools, with your approval before anything important.

Ongoing work: part-time (20 hours a week) from $3,000 a month, or full-time (40 hours a week) from $6,000 a month. Monthly contracts get a lower rate because the hours are guaranteed.

Pay by the hour, or get a fixed price for a project, agreed before I start. AI usage runs on your own OpenAI, Anthropic or Google account, and I estimate the monthly cost before I build.

Pick one job for the AI, and the data it needs.

A written plan: which tools only read, which write and need approval, and how we will know the answers are right.

Tools first, then the model.

Evals on real cases before launch.

Hand-over: code, tests and a short write-up your team can maintain.

shop_mcp_server: an MCP server that gives Claude a store's products, orders and inventory, with one approval-gated write.

ai_shop_assistant: a shopping chat that searches Shopify's live catalog over MCP, so it answers with real products and prices.

product_geo_agent: an AI agent that chooses between five tools, with an eval harness and a live trace.

Three merged pull requests in little_ghost, an open-source Ruby AI agent framework.

More: https://mjesar.github.io/services/ai-in-rails/