I take enterprise AI products from a customer's problem to production. For four and a half years at TEKsystems that meant helpdesk bots: the discovery calls, the integrations, the code, and in the end the roadmap. Now I'm finishing an MBA at NMIMS Mumbai. Next up: AI assistants.enterprise SaaS.cloud platforms.developer tools.
Four long reads and two short ones. Each covers the problem, what I decided, what I built myself, and what came out the other end. Client names are left out on purpose.
An IT helpdesk assistant that files and updates ServiceNow tickets, resets a direct report's password after a one-time code, and finds knowledge articles. I designed how it fit together, from the conversation flows to the integrations behind them.
A new client wanted our IVR password-reset bot but ran Okta and Zendesk, not Azure AD and ServiceNow. I led the rebuild, deployed 70% of the environment myself, and added a dashboard that split bot and human resolution. The client renewed.
Sixteen teams depended on each other, and the rules for those handoffs weren't written down in one place. I interviewed their leads in parallel, wrote 80+ workflows into one web handbook, and had every team signed off on the content in three weeks.
A messy list across 28 states had to become a ranked shortlist the campaign could spend against. I cleaned it, automated the follower counts, and wrote rules that gave a reason for every pick so reviewers could check them in batches.
Seven channels and 10+ explainer projects through one Claude Code entry point: authoring, build, review.
Banking client. Raw data into structured tables on Cloud SQL, 10x faster.
Defined the API requirements for bot-created service tickets at a support org where live agents filed about 80,000 tickets a day. It stayed a proof of concept.
5C analysis and consumer research to find gaps in the chain's offer to customers, then a segmentation, targeting and positioning model.
IoT parking app for corporate campuses: IR sensors, a Flask backend, MySQL, and a live occupancy map.
Most of my work has been getting one system to talk to another inside a client's stack. These are the ones I've written production code or integration requirements for.
My career written the way I'd write a changelog, newest first. + is something I shipped, ~ is something that changed.
Mumbai · --:-- IST
I grew up in Jamshedpur, studied computer science at ITER in Bhubaneswar, and joined TEKsystems as a trainee in January 2020.
Most of my time there went into conversational AI for IT service desks. I started out writing Cloud Functions that talked to ServiceNow. A few years later I was designing how the whole bot fit together, and by the end I owned the roadmap for an eight-person team and was the person the client called.
The work I'm best at sits in between: finding out what a customer actually needs, deciding what gets built first, and making it run inside their stack. Product managers and forward deployed engineers both live in that gap, and I've done both halves of it. I'm doing an MBA at NMIMS to learn the business side properly, so cost, marketing and strategy aren't guesswork. I still write code.
Every sAIge client came with its own requirements. I designed the conversation flows around what they told me about their users and their systems, before anyone wrote code for them.
from sAIgeA new client isn't going to swap Okta for Azure AD to suit a vendor. So the bot changed instead: identity calls rewritten for Okta, tickets moved to Zendesk, and about 70% of that environment deployed by me.
from the password-reset replatformA password-reset bot gets used by someone locked out of their laptop, on a phone call, probably late for something. I designed the voice flows around that person: few steps, a one-time code instead of security questions, and a route to a human when the bot can't finish.
from sAIge and the password-reset botOn the Okta and Zendesk rebuild, the client cared about one thing: how many resets the bot finished without a person. So the dashboard led with bot-versus-human resolution, with ticket counts at each stage underneath.
from the password-reset replatformSixteen teams had to agree on how work moved between them. Every SOP went into markdown and every change went through a GitHub review, so each rule had an owner and a history. The content locked in three weeks.
from the campaign handbookWhen I cut 1,463 influencers down to 273, every pick came with a written reason. Reviewers didn't have to trust a score. They could read the reason, disagree with a specific rule, and say so.
from the influencer shortlistElectives, with the product-facing ones highlighted:
Admitted through CAT 2023 at the 92.43 percentile. View scorecard
On Tata Steel's VG Gopal Scholarship for all four years. Final-year project: IntelliPark.
Secretary of the Interact Club, a Rotary-sponsored student club.
Every card opens the original document or the issuer's verification page.
NMIMS Mumbai
ITER Training and Placement Cell: ran logistics for campus placement drives.
IIT Kharagpur's cultural festival, two seasons in a row.
St. Mary's English High School. Ran intra-school events, from Counter-Strike tournaments to mehendi and chess contests.
The old gallery went offline. A new one is being put together and will live on this site.
A four-year diploma in fine arts, finished in 2014 alongside school.
School events, a college placement cell, two seasons of Spring Fest. I like the side of an event the audience never sees.
27,000+ km behind the wheel and counting.
I'm looking for product manager and forward deployed engineering roles on AI and enterprise products from April 2027, ideally in Hyderabad or Pune, and open to anywhere in India. Email gets the fastest reply.