|| श्री कृष्ण सदा सहायते ||
Darshan KansaraMicrosoft · Redmond

Darshan KansaraSenior Manager, Platform Reliability & OperationsMicrosoft · Redmond

I learned technology where failure is physical.

That lesson still shapes how I operate cloud platforms, improve reliability, work with AI, and build products of my own.

Trace the progression See what I build
physicalserviceclouddataoperationsdecision SYSTEM / SIGNAL / CONSEQUENCE1998NOW
The abstractions changed. The need for evidence did not.

25+ yearsfrom physical infrastructure to AI-enabled operations

Microsoft since 2011datacenter, cloud, continuous improvement, data platforms

Still shippingLifeLens and ApplyReady

01 / The operating model

The technology changed.
The questions accumulated.

A career is not a list of roles. It is a set of instincts earned one layer at a time. Each layer below still informs the one above it.

  1. 1998
    Layer 01

    Infrastructure

    Servers, networks, operating systems

    What is the machine actually doing?
  2. 2005
    Layer 02

    Service operations

    Availability, incidents, customer consequence

    What happens when the service fails?
  3. 2011
    Layer 03

    Datacenter + cloud

    Instrumentation, automation, safe change

    How do we learn before production?
  4. 2020
    Layer 04

    Cloud operations

    Capacity, supply, cost, continuous improvement

    Where is pressure accumulating?
  5. 2022
    Layer 05

    Data platform

    Reliability, security, continuity, economics

    Can everyone see the same system?
  6. Now
    Layer 06

    AI-enabled operations

    Telemetry, runbooks, bounded assistance

    Which decision needs a better signal?
Read the complete career story

02 / The useful contradiction

Operate systems at scale.

Still build from zero.

The distance between those two keeps the judgment honest.

03 / Built, not presented

Two products. One operating principle.

Expose the evidence. Protect the person. Keep the consequential decision in human hands.

Live on Android

LifeLens

A private photo intelligence product for understanding clutter without surrendering the library.

The problem
Photo libraries grow faster than people can understand them.
The boundary
Analysis stays on device. Uncertainty means protect. Deletion requires review.
The build
Flutter, on-device intelligence, Android, English, Hindi, and Gujarati.
Enter the LifeLens product story
LifeLens home screen showing a private photo library LifeLens storage analysis in plain language Real Android product
ApplyReady evidence-led role-fit comparisonReal browser product

Live on the web

ApplyReady

A job-application workspace that asks whether the evidence supports applying before it starts rewriting.

The problem
Most tools optimize the document before testing the fit.
The boundary
No invented experience, no account gate, no stored resume text.
The build
Deterministic browser-side analysis with editable application outputs.
Enter the ApplyReady product story

04 / Operating records

Evidence is what remains after the title is removed.

01

Cloud economics

$1M+annual Azure platform savings
Situation
Recurring infrastructure spend was growing faster than operational understanding.
Intervention
Made cost an engineering signal: capacity, reservations, idle resources, automation, ownership.
Consequence
Reduced annual platform spend while making the economics easier to operate.
02

Reliability

Pre-productionbefore global change
Situation
Infrastructure changes carried uncertainty too close to live Azure datacenters.
Intervention
Established the first Azure Datacenter Instrumentation & Pre-Production Lab.
Consequence
Moved failure into a place designed for testing, instrumentation, and learning.
03

Platform operations

One viewacross engineering and operations
Situation
Reliability, security, privacy, continuity, and cost were managed as separate concerns.
Intervention
Connected telemetry, readiness, ownership, and review around the same platform.
Consequence
Made operational risk visible earlier and accountability less ambiguous.
Open the impact record

05 / The newest layer

AI belongs inside the operating loop, not above it.

Useful operational intelligence starts with telemetry and workflow context. It helps narrow a cause, retrieve a runbook, expose a cost signal, or frame a decision. It does not erase uncertainty or ownership.

See the applied AI model
  1. 01ObserveTelemetry · incidents · cost
  2. 02ContextualizeHistory · ownership · runbooks
  3. 03AssistPatterns · options · uncertainty
  4. 04DecideHuman judgment remains visible

06 / Field notes

Written from inside the system.

Darshan Kansara / Redmond

Systems worth understanding.
Work worth improving.
Ideas worth building.

Start with the real question