AI integration and AI agents for your product
I add AI to software the careful way: the model suggests, your code checks, and people can review and undo anything that matters.
- Pricing
- Estimated per project
- Where
- South Florida or remote
AI agents, assistants and MCP servers I build
Assistants in your app
Chat that can look things up and take actions in your product, limited to what each user is allowed to do.
MCP server development
Let Claude, ChatGPT and other AI tools use your product through secure connections your customers can revoke.
Document extraction
Pull structured data out of PDFs and forms, check it, and let a person confirm it.
Data cleanup
Use AI to fix and fill in messy records, with validation and a plain-code fallback when the model is unsure.
AI agents and automations
Multi-step AI tasks with clear limits, logs and approval points.
Safeguards I've shipped
- Answers are checked before they are sent
- A separate checker reviews the assistant's claims before an answer reaches the user.
- Every AI call has a known cost
- Each call's cost is recorded when it's made, by conversation and by feature, on a costs dashboard. Each user's daily limit is reserved up front so it can't be overrun.
- Spending stops at the budget
- On a platform many businesses share, AI data enrichment runs against a hard budget, and the cost of every outside service is tracked in one place.
- Staff can't see private data by default
- Support staff see a customer's real numbers only after the owner approves a time-limited grant, and every grant is audited.
- AI can't touch secrets or deploy settings
- An AI site editor turns staff requests into code changes in GitHub Actions, with an allow-list that blocks secrets and workflow files, a review gate and a revert workflow.
- A fallback when the model isn't sure
- Image identification runs search and a vision model side by side, with text recognition as a last resort whose confidence is capped.
From my recent work
- An AI assistant that asks before it acts
- In my own personal finance app, I built an AI assistant that looks up the right information before it answers. Its tools can only reach the signed-in household's data, and it can't change anything until the user approves. Every approved change is logged and can be undone for 24 hours.
- A safe way for AI apps to use your data
- In the same app, I built an MCP server, the standard way AI assistants like Claude connect to other apps. People sign in with OAuth and grant read-only or read-and-write access, an admin can block any connected app, and a kill switch shuts it all off.
Common questions
- How is AI integration priced?
- Each project is estimated up front. After we talk through what you want AI to do, I'll send a written estimate, including what it will cost to run. Work is billed hourly in prepaid blocks, and I re-estimate the rest after each one.
- Why not a fixed price?
- AI features need testing and tuning on real data. Billing by the hour lets us adjust between blocks, and you always see where the budget stands.
- Which AI model should we use?
- The one that does your task well enough at a price that works. I test a few on real examples from your product and keep your code able to switch.
- How do you stop AI from doing the wrong thing?
- By never trusting it alone. The model proposes, your server checks permissions and data, and people approve anything important. Changes that matter can be undone.
- What will it cost to run?
- It depends on how often it runs and which model you use. I measure the cost per task during testing and set limits before launch.
- What is an MCP server?
- MCP, the Model Context Protocol, is an open standard that lets AI assistants such as Claude and ChatGPT connect to other software. An MCP server lets your customers connect their assistant to your product, with their permission.
- Can you add AI to an app we already have?
- Yes. I start with one feature where AI clearly helps, build the checks around it, and add more once it proves itself.
- Will you sign an NDA?
- Yes. I'm happy to sign a standard NDA before you share the details.
How we get started
Tell me the idea
What you want AI to do in your product, and who it's for.
I'll ask questions
About your data, your users, and what the AI should and shouldn't be allowed to change. Answering them often sharpens the plan on its own.
Free call
If it looks like a good fit, we flesh out the plan together on a free call. If AI isn't the best answer, or there's a simpler way, I'll tell you.
Written estimate
The scope of work, an estimated cost to build it, and what it will cost to run. We adjust it until it fits your budget and goals.
Work in blocks
You prepay a block of hours, and I start. After each block we review, test and re-estimate the rest.
Tell me what you're working on.
A few sentences is enough. I'll reply within one business day with questions or a next step.