AI automation and agentic systems

I design and build AI automation and agentic systems that do real work in your business.

I'm Tushar Gupta: architect, product lead and AI engineer in one. You work with me from the first question to production, with no hand-offs between a strategist, a project manager and a developer.

~10 yearsbuilding production backends and data pipelines
Recent workproduction AI agents, migrations and database fixes
A live demoyou can try now, on fictional US listings

What I build

Outcomes, not a menu of bots.

Agents that take on recurring work

Software that drafts, sorts, matches and follows up inside the tools you already use. A person approves anything that matters before it goes out.

For example: every inbound inquiry arrives qualified, routed and with a drafted first reply.

Automations between your systems

Data that moves on its own between forms, inboxes, spreadsheets, your CRM and your back office, with checks so bad data stops instead of spreading.

For example: a monthly owner report assembled from data you already hold.

The foundations that keep AI reliable

Backends, data pipelines and databases built the way production systems need to be built: monitored, tested and easy to hand over.

For example: a scheduled pipeline that tells a human when it fails instead of failing silently.

Featured demo · fictional US listings

Buyer request → listing matcher

A buyer writes in plain English: “3 beds and a big backyard for the dogs, 500 to 600k, could stretch to 640k, within 2 miles of Riverside Park.” The system returns ranked listings, a “why it matched” breakdown for each, a lead score, follow-up questions for anything missing, and a CRM-ready record.

  • Firm requirements (budget, bedrooms, must-haves) are hard filters. A listing over budget can never appear as a match.
  • Near misses are shown separately, with the one requirement they miss.
  • The architecture is explained on the page, including where an LLM fits and where it is not allowed to decide.

This is a demo I built, not client work. Every listing, neighborhood, school and price is invented. It runs entirely in your browser.

Screenshot of the listing matcher demo: a buyer message, a lead score card and the top matching fictional listing with the filters it passed.
Screenshot of the demo. Fictional US listings.

How working with me starts

The first payment is a working proof of concept, not a call.

  1. Free discovery

    Free · optional

    A 30-minute call, or stay on email if you prefer. I learn where your week goes and tell you honestly whether AI is the right fix.

  2. Paid proof of concept

    From $200

    A small, time-boxed slice of your real workflow, working. The fee is credited to the build if you continue.

  3. Fixed-scope build

    50% to start · 50% on delivery

    A written scope and a fixed price for the rest, agreed after the proof of concept. Everything is built in your accounts.

  4. Two months of maintenance

    Included

    Bug fixes and breakages are covered for two months after go-live (new features are quoted separately). I acknowledge issues within one business day.

How I build

Engineered like production software, because it is.

Most AI projects don't fail on the prompt. They fail on the plumbing: missing context, no tests, no limits, nobody watching. I test every system on your own examples before go-live and show you where it fails, not only where it works. People approve the steps that matter, costs are capped, and every action is logged.

Read the technical approach →

About me

One accountable person, end to end.

I've spent about ten years as a software engineer and architect, leading work on high-traffic production systems: APIs, data pipelines, databases and cloud infrastructure in Python, Go and Java. Today I focus on AI agents and automation for small and mid-sized businesses.

When you hire me, the person who scopes the work is the person who builds it and the person who fixes it. You own the code, the accounts and the data.

Hours. I'm in India (IST) and work async first: progress lands while you sleep, and replies come within 24 hours on weekdays.

Calls are Monday to Friday, 8–10 AM or 6–9 PM IST. In US winter time the evening slot is 7:30–10:30 AM Eastern, and the morning slot is 6:30–8:30 PM Pacific the day before. Add an hour during US daylight saving time.

LinkedIn profile

What I won't do

Saying no is part of the job.

  • Invent metrics or show a demo as if it were client work.
  • Promise that “AI runs your business”. People stay in charge of the decisions that matter.
  • Gray-area scraping, or anything that breaks a site's terms.
  • Trading, betting or get-rich bots.
  • Sell you AI you don't need. If a spreadsheet rule or a plain script solves it, I'll say so.

FAQ

Questions people ask first

Why is discovery free and optional?

You shouldn't have to pay a stranger for a conversation. Thirty minutes is enough for me to understand the problem and for you to decide if I'm the right person. If you'd rather not get on a call, send the details with the form below; that works just as well.

What is the paid proof of concept?

A small, time-boxed piece of your real workflow, built and working, usually on a sample of your own data. It starts from $200, with a fixed price agreed before I begin, and the fee is credited to the full build if you continue. If it doesn’t prove the idea, you stop there with no further commitment.

What do the two months of maintenance cover?

Bug fixes and anything that breaks in what I delivered, for two months after go-live, at no extra charge. New features and scope changes are quoted separately. I acknowledge every issue within one business day.

Who owns the code and the data?

You do. I build in your accounts (your cloud, your automation tools, your repositories) wherever possible, and hand over the code, the credentials and plain-English notes on how it works.

How does async work across time zones?

You send a short brief, I reply with questions or a plan, and progress lands overnight with a written update. Calls happen in the overlap windows when they help. Most decisions can be made in writing.

Contact

Tell me the work that eats your week.

Two lines are enough: what the task is and how often it happens. I reply within 24 hours on weekdays, with honest next steps, even if the answer is that you don't need AI.

Send me your requirement

The form is delivered to my inbox by FormSubmit. See privacy.