Sourodiptto Mondal

I build products where complexity becomes clarity.

Sourodiptto MondalFounding Product Manager, La-Fetch. Gurugram, India.
A notebook. Still being written.

the thing you’re actually building. it’s rarely the feature.

Before building the solution, I try to understand the system.

a problem is what you notice. the system is what’s producing it.

A request usually arrives as a feature. Underneath it there is nearly always a system: people behaving in ways that make sense to them, a business with real constraints, data that answers only the questions you think to ask, technology with its own costs, and an operation that decides what actually gets delivered.

I’m comfortable when that picture is incomplete. I start from first principles, work out which parts matter, and turn what I learn into decisions a team can act on.

The loop, and the walls around itA loop of four steps — watch behaviour, interpret it, decide, change the product — enclosed by three walls: business, technology and operations.business — what is worth doingtechnology — what is possibleoperations — what is deliverablewatchwhat people dointerpretwhat it meansdecidewhat to changechangethe productone turn
fig. 1 — the loop, and the walls around it.

the walls move — slowly, and usually when someone pushes.

  1. 01 — ambiguity

    Not missing information — information nobody has organised yet. The first useful thing is usually a better question.

  2. 02 — behaviour

    People are consistent; just not always with what they told you. Watch the funnel first, then go and ask.

  3. 03 — systems

    Every product is also an operation. Somebody packs it, somebody gets paid for it, somebody answers when it breaks.

  4. 04 — constraints

    A constraint is a fact about the world, not a failure of imagination. Design starts where it stops being negotiable.

  5. 05 — feedback

    Nothing improves without a loop. Half of product work is shortening the distance between what happened and who needs to know.

The whole thing, in six plates.

hover, tap or tab through them

Who is this for, and what are they actually trying to do?

Everything downstream inherits this answer. Get it wrong and every later metric is a very precise measurement of the wrong thing.

where I’ve touched it

  • Zing Read customer behaviour and the operational friction around it.

Evidence, roughly in the order it happened.

(not the order that looks best)

Zing · Founder’s Office · Sep 2022 – Apr 2024

From pre-order to bulk delivery

how it reached people

1. observation
Customer behaviour, and the operational friction around it.
2. intervention
Informed a strategic shift from pre-order to bulk delivery. Also helped shape the early business model and go-to-market direction, from zero.
3. outcome
Fulfilment efficiency and adoption improved.
the full note

23% increase in sales

Perfect Homz · Growth Associate · Jun 2024 – Jan 2025

1. observation
Growth opportunities, found through performance and market analysis.
2. intervention
Shaped targeted acquisition initiatives.
3. outcome
Contributed to a 23% increase in sales.
the full note

17% lower marketing cost, quarter over quarter

1. intervention
Built MySQL-driven customer segmentation to refine campaign targeting.
2. outcome
Marketing costs fell 17%, quarter over quarter.
the full note

Pallete (Twinleaves) · Associate Product Manager, internship · Jan – May 2025

No number on this page. Funnel drop-offs and behavioural patterns, read in Mixpanel, turned into prioritised iterations. PRDs. Feedback loops set up between engineering and design, so there was less ambiguity during API and integration work.

On the desk.

The things I actually touch, redrawn from habit. Illustrative — none of it comes from any company’s documents. They move; it’s a desk.

PRD · draft 0.3
  1. 1 Problem
  2. 2 Hypothesis
  3. 3 Success looks like
  4. 4 Non-goals
  5. 5 Open questions
write the non-goals first.
funnel · illustrative

arrive

look

try

commit

return

where I’d look first
how the numbers relate
revenueordersbasket sizevisitors××conversiontrust · speed · price
a hypothesis, before it’s a feature

We believe that for will . We’ll know we’re right when moves.

+ what would change my mind
roadmap fragment

now

next

later

a flow, with one decision
landchoosesure?payno
the data underneath

user 1 — ∞ order 1 — ∞ item

Then: a storefront that didn’t exist yet.

La-Fetch. June 2025.

Monthly revenue up 12.5 times, from ₹40K to ₹5L

1. problem
Joined before launch. The storefront wasn’t built; the codebase belonged to an agency.
2. outcome
Monthly revenue, ₹40K → ₹5L. Weekly orders, ~10 → ~75, at ₹1,500 a basket. Peak: over 20 orders in a day.
the full note

39% of orders use partial cash on delivery

1. problem
Cash-on-delivery orders were dropping off.
2. intervention
Shipped partial-COD and EMI checkout, through Razorpay and GoKwik.
3. outcome
Partial COD is now 39% of all orders, and contributed to a 7× rise in weekly order volume.
the full note

two systems, sketched —

8,000 plus

SKUs, tagged across seven attributes — by a system that had to know when it wasn’t sure.

  1. it reads

    • product description
    • existing tags
    • images

  2. two models

    • model
    • model

    consensus

  3. and scores itself

    • high
    • medium
    • low

  4. high

    tag approved automatically

    uncertain

    ~1,500 SKUs go to a person

  5. it makes

    • the attribute tree

    which became the site’s navigation

the part I’d defend: it can say “I’m not sure.”
fig. 2 — how a catalogue learns what it doesn’t know. the full note

A support assistant that isn’t allowed to improvise.

during the conversation

  1. the customer taps a button
  2. order tracking · cancellations · returns · refunds
  3. an answer no model wrote
  • free-text input
  • LLM-written replies

no hallucinated policy answers. no prompt-injection risk.

after it ends

  1. the conversation closes
  2. Claude API — summarise
  3. a summary
  • summariser down? a fail-safe path — the live chat never notices

the model only comes in here — after, never during.

fig. 3 — deciding what a model may touch. the full note

Not every product decision is a feature. One of mine was a team.

  • Established the in-house engineering function.
  • Migrated the full codebase off external agencies.
  • Owned the first sale event end to end — load balancing, network stabilisation, live sprint triage.
  • Set up AWS staging, deployment and network security from scratch.
the full note

The record.

Four roles, oldest first. Read down and the type gets larger.

  1. Sep 2022 to Apr 2024
    Bengaluru

    Product Strategy & Operations Associate

    Zing, Founder’s Office

    Read customer behaviour and operational friction to inform a shift from pre-order to bulk delivery, and helped shape the early business model and go-to-market direction.

    looking at: businesses

  2. Jun 2024 to Jan 2025
    Bengaluru

    Growth Associate

    Perfect Homz

    Found growth opportunities through performance and market analysis, and built the segmentation that cut marketing costs.

    looking at: customers, and growth

  3. Jan 2025 to May 2025
    Bengaluru

    Associate Product Manager

    Pallete (Twinleaves), internship

    Read funnels in Mixpanel for product opportunities; wrote PRDs; set up feedback loops between engineering and design.

    looking at: the product

  4. Jun 2025 to Present
    Gurugram

    Founding Product Manager

    La-Fetch

    Joined a D2C fashion marketplace before launch. Owns product end to end: revenue from ₹40K to ₹5L a month, an in-house engineering function, two production AI systems.

    looking at: systems

B.Tech., Electrical & Electronics Engineering — Presidency University, Bengaluru, India, 2024. Technically, an electrical engineer.

Toolkit, indexed.

What each thing was actually used for — where I can say.

product

  • Product Strategy
  • 0→1 ExecutionZing12.5×
  • Product Analyticsfunnels, behaviourPallete
  • Funnel Optimisationdrop-offPallete39%
  • Experimentation
  • Retention
  • Stakeholder Management

data & ai

platform & tools

  • AWSstaging, deployment, network securityin-house
  • Razorpaypartial COD, EMI39%
  • GoKwikpartial COD, EMI39%
  • Figma
  • Jira
  • Tableau

Currently thinking about

  • systemsWhy does something that works perfectly in the demo fall over on the busiest day?
  • AI productsWhich parts of a product should be allowed to improvise — and which absolutely shouldn’t?
  • human behaviourWhy do people abandon the thing they said they wanted?
  • technologyWhat became cheap recently, and what did that change about what’s worth building?
  • IndiaWhat does a good checkout look like when trust is scarcer than attention?
  • learningWhat is the fastest honest way to find out I’m wrong?
  • buildingWhat’s the smallest thing that would tell me the truth?

(asking “why” twice is usually enough. three times is a hobby.)