Engagement Manager at Trail, formerly Nanonets. Ex-founder, ex-PM.
I get enterprise AI agents working in production.
Once a deal is signed, I own it: I build the AI agents, get them live, prove the ROI and grow the account.
AfterAgents do the routine work. People handle the exceptions.
So far
Built from zero, more than once
~$1M
A US healthcare vertical, built from zero
Revenue in its first year, as the single post-sales owner from discovery to expansion.
8workflows in production
NanonetsTrail2025–now
First
Delhi NCR's first quick commerce for fashion
Fashion delivered in 100 minutes, built from the ground up as one of three founders.
$250Kpre-seed raised
Zulu Club2024–25
~50%
LLM products that became the main growth channel
Share of website traffic from the zero-to-one LLM tools launched.
2×marketing-qualified leads
Nanonets2024
$100K
A year of model costs cut
Through model routing, usage controls, pricing gates and vendor optimisation.
−50%model and API spend
Nanonets2024
How I build
Built to repeat, easy to tailor
I build agents from steps that snap together. Most steps carry over to every customer. The few that differ, like where work arrives or which system it lands in, swap out, and each customer's rules are tuned on top.
Swipe, then tap an agent to see how it's built.
Pick any agent to see how it's built.
Customer order agent
- Reads order emailsswaps per customer
- Finds the right product
- Checks stock and shelf life
- Places the order in SAPswaps per customer
- Confirms to the customer
Tuned for this customer
- Minimum order only on the first order each month
- Only stock with 12+ months shelf life
- Orders after 3pm ship next day
- Same for every customer
- Swaps per customer
- Rules tuned per customer
One deployment, step by step
From entering orders to managing rules
A live project at a Fortune 500 company, where I led the deployment. Their order teams in several countries took orders by email and keyed them into SAP. Now an agent takes each order from the inbox to SAP.
The problem · 1 of 12
Every order arrives as an email
Orders arrive by email, mixed in with follow-ups, changes, questions and re-sends.
How I work
One owner, from first workshop to expansion
What I do
I sit with the people who do the work and map every step, including the messy parts no document mentions.
What you get
A one-page map of how the work really happens today.
What I believe about AI in production
I'd rather solve 80% of cases 100% of the time than 100% of cases 80% of the time
- Shipping an agent is an input. Less work for the customer's team is the output.
- Use AI to build. Keep the thinking for yourself.
- Always probe. The problem you're handed is usually one piece of a bigger one.
- SOPs aren't sacred. A small change to the process can let one agent solve three problems.
- Exceptions go to people. That's not the agent failing; that's the design.
- Make it as deterministic as you can. Let the model choose the tool, not do every step.
- Give each task only the rules it needs, when it needs them. One giant prompt breaks.
- Process and data decide whether an agent works. The model is mostly a question of size.
- Most workflows are a good road with a few broken stretches. AI is the patch, not the road.
- Count ROI in hours freed and revenue gained, never in agents deployed.
— Shivang
Career
How I got here
- 2025NanonetsTrail
Engagement Manager
Owns enterprise AI accounts from signed deal to live agents, ROI and expansion
- 2024Zulu Club
Founding member
Product and growth for a 100-minute fashion delivery startup
- 2024Nanonets
Product Manager, Growth
Zero-to-one LLM document products and the growth engine around them
- 2022MakeMyTrip
Product Analyst
Supplier Deals redesign and post-merger analytics
- 2021Fractal Analytics
Data engineering & science
NLP pipelines and MLOps for a global consumer-goods company
Contact
Say hello
Getting AI agents into production at your company, or hiring for it? Happy to compare notes.