Nilansh Netan Cloud & AI New Delhi

The model is rarely the bottleneck. The plumbing is.

I build at the spot where cloud architecture meets generative AI.

Welcome inside. This is where the real work happens.

SCROLL TO ENTER
How I think

I'm a generalist by choice. Depth is easy to admire, but range is what connects the dots. I explain systems through analogy, because that's how understanding actually travels. Five years in New York taught me rigor. Coming home to Delhi gave me double vision: I see this place as someone who grew up here, and as someone who left. I build with AI, and I think with it too.

My story

My life through this Dino run…

Scroll ↓ to see my run so far

01 / 13
SCROLL TO RUN
  1. lvl 01 / 13 — new-delhi.init

    Where it all starts

    New Delhi. First computer, endless curiosity, zero idea how far the run would go.

  2. lvl 02 / 13 — bachelors-cs

    Four years of computer science

    The fundamentals — systems, networks, algorithms. Learning how things actually work under the hood.

  3. lvl 03 / 13 — hanu.intern

    First badge: intern at Hanu

    First production codebase, first real cloud. The gap between tutorials and prod, crossed the hard way.

  4. lvl 04 / 13 — cloud-engineer

    The return offer

    Cloud Engineer at Hanu. Two years and four months of shipping on Azure — and earning the keys to production.

  5. lvl 05 / 13 — nyu.admit

    New York says yes

    An admit from NYU. A life packed into two suitcases, pointed at JFK.

  6. lvl 06 / 13 — nyc.spawn

    New map: New York City

    New city, new speed. The five New York years begin.

  7. lvl 07 / 13 — ms-cs

    M.S. in Computer Science

    Grad school at NYU — going deep on the theory underneath everything I'd been shipping.

  8. lvl 08 / 13 — vital-cloud-labs

    SWE at Vital Cloud Labs

    Part-time software engineer in the CSE department under Prof. Reddington — building cloud tooling between lectures.

  9. lvl 09 / 13 — adobe.sre

    Site reliability at Adobe

    An SRE internship at real scale. A pager teaches faster than any textbook.

  10. lvl 10 / 13 — back-to-lab

    Back to the lab

    Returned to Vital Cloud Labs to finish what I started, shipping until graduation.

  11. lvl 11 / 13 — ms-cs.grad

    Graduating with an M.S. in Computer Science

    Cap and gown in NYU violet. The degree earned between lab shifts and pager duty.

  12. lvl 12 / 13 — ensemble

    Senior Engineer, Cloud & AI

    Three years at Ensemble Health Partners, working the exact intersection this site is about: cloud architecture meets generative AI.

  13. lvl 13 / 13 — h1b.respawn

    The one obstacle I couldn't jump

    The H1B lottery didn't pick me. So: respawn in New Delhi — same team, now as part of Ensemble Global. Same runner, new map. This chapter is still being written.

Work, told properly

Not a project list. What actually happened.

01 / 04
case/01 — rag-at-scaleRAG · Azure AI Search · Azure OpenAI

Teaching an enterprise to talk to its own documents

The problem
[200+ SOP documents nobody could find anything in. People asked colleagues instead of searching.]
What I noticed
[The hard part wasn't the model. It was retrieval quality and trust. Three wrong answers and nobody asks a fourth question.]
What shipped
[A RAG pipeline scaled to the full document set. Add a metric: queries per week, time saved, adoption.]
case/02 — meet-them-in-teamsBot Framework · Teams · Azure OpenAI

A chatbot that lives where people already work

The problem
[Another portal nobody opens? No. Support had to meet people inside Teams, where questions were already being asked.]
What I noticed
[Fallback design mattered more than the happy path. Knowing when to hand off or say "I don't know" made it production-grade.]
What shipped
[A Teams-native chatbot with custom question answering and OpenAI-powered validation, deployed org-wide.]
case/03 — past-factory-settingsCustom MCP server

Extending AI tools past their factory settings

The problem
[Off-the-shelf AI clients couldn't reach the systems that mattered. The interesting data lived behind internal APIs.]
What I noticed
[MCP flips the integration problem. Build the server once, every compatible client gets the capability.]
What shipped
[A custom MCP server plugging real systems into modern AI clients. Name the workflow it unlocked.]
case/04 — coherent-past-msg-oneCopilot Studio · Session management

Keeping agents coherent past message one

The problem
[Agents in Teams forgot everything between turns. Multi-step tasks fell apart the moment context was needed.]
What I noticed
[State is the unglamorous half of agent design. Session management separates a demo from a tool.]
What shipped
[Session-management infrastructure keeping multi-turn conversations coherent at enterprise scale.]
Now

What has my attention.

UPDATED JULY 2026
Ideas

Things I keep saying.

Vendor risk should work like bloodwork, not an annual physical. Continuous monitoring catches what a point-in-time audit never will.

On third-party risk

Classical architecture solved agent problems centuries ago. Load-bearing structure, redundancy, graceful failure. We keep rediscovering it.

On agent design

The fastest way to make a complex system click is to compare it to something you already understand. Analogy is a load-bearing tool.

On explaining things

More on LinkedIn →

Off hours

Away from a terminal you'll find me cooking something I'll definitely over-season, reading market charts like other people read fiction, and planning the next trip before the last one is unpacked.

Contact

I reply to interesting problems.

Building something at the cloud and AI intersection, or want a second brain on an architecture? Say hi.

Email me LinkedIn GitHub