Skip to Content

AI Agents & Automation

AI agents that take action, not just chat

A chatbot answers questions. An agent does the next step: checks a calendar, creates the CRM record, drafts the email and waits for your approval before sending. I build agents around one clear job in your business, wired to the tools you already pay for.

Updated

Problems this solves

  • Your chatbot answers questions, then a person still has to do the actual work afterward.
  • Leads arrive from several places and nobody has time to qualify, log and follow up on each one the same day.
  • You tried a general AI tool, but it can't touch your calendar, CRM or inbox, so it stays a toy.
  • You worry an AI that can send emails or edit records will do something wrong with nobody watching.

What I build

  • An agent that reads an inquiry, checks availability and books the appointment in your calendar
  • A lead agent that qualifies a new contact, creates the CRM record and drafts a personal reply
  • An inbox agent that sorts incoming email, drafts replies and flags what needs a person
  • An internal agent your team asks in plain language to look something up or update a record
  • Approval steps, so risky actions wait for your yes before they run
  • A log of every action the agent took, so you can see and correct what it did

How it works

  1. 1Pick one job

    We choose a single task the agent should own and write down what done looks like. Small and clear beats ambitious and vague.

  2. 2Connect your tools

    I wire the agent to your calendar, CRM, email or sheets through their official APIs, with the narrowest access that works.

  3. 3Test with real cases

    I run it against real examples from your business, tighten the rules and add approval steps where a mistake would cost you.

  4. 4Launch and watch

    It goes live with a log of its actions. We review the first runs together and adjust before you rely on it.

Work that shows it

Tools I use

  • Claude, OpenAI and Gemini APIs
  • Tool calling
  • Next.js
  • TypeScript
  • Python
  • Supabase
  • Google Calendar and Gmail APIs
  • Webhooks

Working with clients abroad

I work remotely with clients worldwide from Assam, India, and in person across Assam and Guwahati. Quotes are in USD and paid by Wise or international bank transfer. Code lives in your GitHub from day one, and you get a staging link to try at every step.

Questions

A chatbot talks. An agent also acts: it can look things up, create records, book slots and draft messages using your tools. Many projects use both, with the chat on the front and the agent behind it.
Only for low-risk steps. For anything like sending an email or changing a customer record, I add an approval step so you confirm first, and you can loosen it later once you trust the results.
Yes. I work remotely with clients abroad over video calls and shared documents, and quote in USD. Payment is by Wise or international bank transfer, and I plan calls around your time zone.
You do. The agent runs on accounts opened in your name, and you receive the source code and instructions. I don't lock anything behind my own subscriptions.
It depends on the tools involved and how many cases the agent must handle. You get an estimate after a first call, and a working version to try early rather than at the end.

Tell me what you have in mind, and I'll tell you how I'd build it.