AI agent development for your business
I design, build and test AI agents that do a real job in your business: answering customers, qualifying leads, preparing reports, checking documents. They are built around your tools and your people, not a generic chatbot.
What you get
- A workflow map showing exactly what the agent will and will not do
- A system prompt with guardrails, written like a contract and kept under version control
- Tools connected to your systems: APIs, an MCP server, spreadsheets, WhatsApp or the browser
- An evaluation set built from your real cases, and a test report before launch
- Human approval and escalation rules for anything risky
- Logs so you can review every conversation and action
- Handover notes and a call after launch
How I make agents reliable
AI models make mistakes, so reliability comes from design rather than hope.
- Narrow jobs: one agent, one clear responsibility
- Least-privilege tools: read-only first, write access only where needed
- Approval gates on anything that sends, pays or deletes
- Scenario tests that include bad inputs, Tamil and mixed-language messages and attempts to confuse the agent
- A verifier step that reproduces a failure before anyone is told about it, the same approach I use in MyTest
- Monitoring after launch, with weekly reviews
Examples of agents I design
- Shift analyst for a spinning mill: reads production and downtime logs and drafts the shift report in Tamil and English
- Export desk for a textile exporter: turns buyer emails into quotation drafts
- Enquiry agent for a wedding venue: answers in minutes, books site visits, logs every lead
- Front desk for a clinic: books appointments and collects intake forms, never gives medical advice
- After-sale concierge for a jewellery showroom: thank-yous, review requests and scheme reminders on WhatsApp
How a project works
Scoping call
We pick one workflow worth automating and define what success looks like.
Prompt and tools
I write the agent's instructions and connect the tools it needs.
Test on real cases
We run your real examples and edge cases and fix what breaks.
Launch with a human in the loop
The agent starts with approvals on, and we loosen them as trust grows.
Review and improve
Weekly look at the logs, prompt fixes and new cases.
What I need from you
- One real workflow that costs your team time today
- Sample data: messages, documents or reports the agent will handle
- A person who knows the work, for a few short calls
- A clear list of things the agent must never do
Technology
I use leading AI models such as Claude through their APIs, tool calling, MCP servers, retrieval (RAG) when answers must come from your own documents, and TypeScript and Python for the code. I choose the simplest approach that works, because simple agents are easier to trust and cheaper to run.
Frequently asked questions
What is an AI agent?
An AI agent is software that uses an AI model to complete a task: it reads input, decides what to do, uses tools such as your calendar, CRM or WhatsApp, and hands over to a person when it is unsure. A chatbot mostly answers questions, while an agent does the work.
Which AI model do you use?
It depends on the task, the language, the cost and your data rules. I often use Claude, and I can use other leading models through their APIs. I pick the model after testing on your real examples.
What happens if the agent makes a mistake?
Mistakes are planned for. Risky actions need human approval, every action is logged, and the agent hands over when it is unsure. After launch we review the logs weekly and fix the prompt or tools.
Can you add an agent to my existing website or app?
Yes. I can embed an agent in your website or app, and I built an embeddable agent SDK, Vanta Embed Agent, for exactly this.
What does AI agent development cost?
The cost depends on the number of tasks, integrations, channels and review steps, plus running costs for the AI model and messaging. The cost guide explains each factor, and I give a written estimate after the free first consultation.
Keep reading
Tell me what eats your week.
My intake agent asks a few quick questions and puts your brief in front of me. I read every one.