Meet Soni

Building AI.
Building businesses.

AI product builder, AI venture strategist and GTM & growth operator.

Explore the work

Products and the
market around them.

I build AI-powered products and connect technical decisions with the people, problems and markets they serve.

About Meet

About

I build AI-powered products and connect product decisions with customer problems, positioning and business economics. The work here includes a local-first founder tool, a conversational beta and research.

On the product side, I work on workflows, rules and reviewable actions. On the venture and GTM side, I examine who the buyer is, what they use today and which assumptions need a real-world test.

The through-line is an operator's view of software: it should give time back, not add another surface to manage.

01

Build from evidence

Research and unit economics come before a build. If the numbers do not support the thesis, the answer is no.

02

Ship the real thing

Prototypes should answer a question. Products should change how someone works.

03

Respect the operator

Software earns its place by giving time back, not by adding another surface to manage.

Intelligent systems,
built to be useful.

From conversational agents to founder workflows, I turn a business problem into a product that can be tested.

View selected work
Explore the product lab

AI Lab

Working prototypes, a free beta and a GTM framework. Open a case study to see the contribution, decisions and current evidence.

AI Chief of StaffWorking prototype

FounderPilot

A local-first desktop prototype that ranks attention items and puts suggested actions through founder review.

Read the case study
Conversational operationsFree beta

WhatsApp AI Agent

Free beta testing by request, focused on qualification, responses and human handoff. Channel coverage is being validated.

Read the case study
GTM strategy engineExperiment

AI Venture & GTM Lab

A structured loop from market discovery to retention for early AI ventures.

Read the case study

Status labels reflect the current state of each build. Planned capabilities are not presented as shipped.

FounderPilot

A local-first AI Chief of Staff for small-business founders. Find what deserves attention and decide what happens next.

Working desktop prototype. Ranked attention, suggested actions and founder-controlled decisions.

Read the FounderPilot case study

Working prototype, AI Chief of Staff

FounderPilot

Problem and why it matters

A small-business founder needs to decide which overdue invoice, lead or task deserves attention without rebuilding the picture from separate records each morning.

A useful assistant must explain its priorities and leave commercial decisions with the founder.

My contribution

Built a local-first Windows desktop prototype with a dashboard, CSV imports, attention ranking and a reviewable action queue.

Approach

  • Parse and validate local lead, invoice and task CSVs before applying imported records.
  • Use deterministic rules and a configurable playbook to rank up to five attention items and explain their triggers.
  • Keep the default workflow local. Optional AI assistance explains or drafts; it does not control ranking.

Decisions

  • Separate suggestions from execution: the founder can approve, edit, skip or snooze a proposal.
  • Protect workspace records with encryption at rest. Browser memory notes are a separate, unencrypted local feature.

Implemented and available now

  • Local dashboard and CSV validation, rule-based ranking, copy-ready drafts, task handling and an encrypted workspace.
  • Optional calendar and AI integrations have separate setup and consent boundaries.

The founder workflow

  1. Import and review records
  2. Rank attention items with reasons
  3. Review a suggested action
  4. Approve, edit, skip or snooze; use the resulting local draft

Evidence

  • The local implementation contains CSV validation, deterministic ranking and approval-queue modules, with documented data boundaries.
  • Approvals create local drafts or local task entries. Any recovered amount entered by a founder is self-reported, not independently verified revenue.

Current outcome

A working local prototype under iteration. No customer adoption, time saving or revenue uplift is claimed here.

Takeaway

Explanations and a clear approval boundary make decision support inspectable before adding more automation.

Tools and methods: Electron / Node.js / CSV validation / Deterministic rules / Encrypted local storage

WhatsApp
AI Agent

Conversational automation for lead qualification, response generation and human handoff, now in beta testing.

Free beta testing by request. Channel coverage and feature availability are confirmed before a test.

Explore the beta scope

Free beta, Conversational operations for small teams

WhatsApp AI Agent

Problem and why it matters

Small teams need to turn inbound conversations into useful qualification information while keeping a person available for exceptions.

Response quality, lead context and a reliable handoff matter more than the number of automated replies.

My contribution

Developing the conversation design and beta scope around lead qualification, response generation and human handoff.

Approach

  • Define the information needed to qualify a lead and the boundaries of an acceptable response.
  • Test response generation and escalation against realistic conversation scenarios.

Decisions

  • Keep uncertain or exceptional conversations on a human-review path.
  • Confirm channel coverage and setup with each beta participant before a live-workflow test.

Implemented and available now

  • Free beta testing is available by request. The exact enabled features and channel setup are confirmed when arranging a test.

Scope under validation

  • Lead qualification and response quality
  • Website-chat entry and live WhatsApp channel coverage
  • Human handoff, error handling and production reliability

Evidence

  • The beta invitation is available now. A public implementation walkthrough and measured beta results are not yet published.

Current outcome

An opportunity to test the current beta for free, with feature availability agreed in advance.

Takeaway

Evaluate a conversational system through qualification accuracy, response boundaries and the handoff experience.

Tools and methods: Conversation design / Intent routing / Response generation in beta

Ask about free beta testing

From product
to market.

Discover, validate, position, price, launch and grow. Commercial strategy belongs alongside the build.

The bars are illustrative, not actual client revenue data or a guaranteed outcome.

Explore the venture and GTM work

Experiment, GTM strategy for early AI ventures

AI Venture & GTM Lab

Problem and why it matters

Early AI ventures often build before they understand who pays, why now, and how the product reaches the market. Enthusiasm substitutes for evidence.

A technical prototype and a commercially viable business answer different questions.

My contribution

Structured a seven-stage evaluation framework connecting customer discovery, product choices and commercial assumptions.

Approach

  • Define the market, buyer, costly problem and alternative before selecting a narrow ICP.
  • Write positioning and pricing hypotheses, then specify evidence needed from validation, distribution and retention tests.

Decisions

  • Treat each stage's signal as evidence to seek, rather than proof that an experiment has already succeeded.
  • Use explicit stop or revise criteria before increasing investment.

Implemented and available now

  • A documented discovery-to-retention framework and evidence criteria, shown below.

Evidence

  • Seven stages with a stated question and progression signal. Research projects provide separate examples of economic analysis.

Current outcome

An experimentation framework. Successful launches, paid customers and retention outcomes are not claimed.

Takeaway

A growth decision is stronger when the assumption, test and evidence threshold are stated separately.

Tools and methods: Market discovery / ICP definition / Positioning / Pricing hypotheses / Experiment design

AI Startup GTM Lab

Discovery, ICP, positioning, pricing, validation, distribution and retention form one test framework. The signals below are evidence to seek, rather than results already achieved.

  1. 01

    Market discovery

    Start from a costly, recurring problem with a named buyer — not from a model capability. Interview before you build.

    Evidence to seek: You can name the buyer, the trigger event and the cost of inaction.

  2. 02

    Ideal customer profile

    Narrow to the segment where the problem is acute, the budget exists and the buying process is short enough to learn from.

    Evidence to seek: A shortlist of accounts you can reach directly this month.

  3. 03

    Positioning

    State the category, the alternative and the reason to switch. Positioning is a decision about what you are not.

    Evidence to seek: A prospect can repeat your value in one sentence.

  4. 04

    Pricing

    Price against the value of the outcome, then test willingness to pay before scaling acquisition.

    Evidence to seek: Real buyers commit at the proposed price, not a hypothetical one.

  5. 05

    Validation

    Run the smallest experiment that can falsify the thesis. Prefer evidence from behaviour over opinion.

    Evidence to seek: A repeatable path from first contact to paid use.

  6. 06

    Distribution

    Pick one channel you can compound in. Depth beats breadth until the loop is proven.

    Evidence to seek: A channel that produces qualified conversations without heroics.

  7. 07

    Retention

    Measure whether the product becomes part of the customer's routine. Retention is the real product-market fit test.

    Evidence to seek: Usage that persists without reminders.

Research before
the next decision.

Unit economics, return behaviour and profitability models help test the business case before committing to build.

Synthetic-data analysisMyntra return & assortment profitability

ResearchQuick-commerce unit economics

Read the research and operating background

Strategy & research

Two kinds of evidence: published economic research and a clearly labelled synthetic-data exercise.

SSRN research

Quick-commerce unit economics

When can Tier-2 expansion improve Indian quick-commerce unit economics?

My contribution

Independent research comparing Tier-1 and Tier-2 revenue drivers, costs, contribution margins and breakeven.

Method and evidence

Comparative unit-economics analysis using public secondary sources.

Tier-1
Higher competition, marketing intensity and operating complexity.
Tier-2
Lower costs, offset by smaller baskets, weaker monetisation and less stable demand density.

Findings and interpretation

  • Lower operating and acquisition costs do not automatically make Tier-2 expansion profitable.
  • Demand concentration, controlled delivery costs and repeat engagement determine whether the cost advantage translates into stronger economics.

Takeaway

Evaluate operating conditions and downside assumptions before treating a city tier as a growth thesis.

Research findings, not evidence of operating a quick-commerce business. SSRN posting does not establish peer review.

Read the original research on SSRN
Synthetic-data analysis

Myntra returns & profitability

How do returns and subsidy costs affect the interpretation of assortment profitability in a synthetic fashion-commerce dataset?

My contribution

Analysed return behaviour, assortment and modeled subsidy leakage across approximately 119K synthetic records.

Method and evidence

Synthetic-data modeling. The figures below are the supplied project summary; the underlying workbook and toolchain are not published here.

Synthetic dataset
Approximately 119K rows
Modeled return rate
Approximately 24.44%
Modeled subsidy leakage
Approximately ₹35.7 lakh

Findings and interpretation

  • Keep the modeled return rate and subsidy leakage separate from any claim about official company performance.
  • Read the results as an analytical exercise in returns and assortment economics, rather than demonstrated cost savings.

Takeaway

A summary metric needs its dataset definition, assumptions and cost context before it can support a business decision.

The Myntra dataset used in this analysis was synthetic. It is not official internal company data and should not be read as such.

Experience

Research, commercial modeling and decision support in a confidential real-estate advisory setting.

Founder's Office

Real-estate consulting firm (confidential)

Internship

Supported leadership with strategic research, commercial models and go-to-market decision materials in a confidential advisory setting.

  • Prepared research dossiers, financial scenarios and sensitivity analysis to support review of commercial assumptions.
  • Contributed to GTM strategy materials and reusable research and AI-workflow documentation.
  • My contribution was analysis and decision support. Client relationships, final investment decisions and commercial outcomes belonged to the wider team.

The employer is kept confidential. No client information or commercial transaction details are disclosed. The work described here concerns my analytical contribution, without attributing team deals or financial outcomes to me.

Let’s build
what’s next.

Building an AI product or hiring across product, strategy and growth? Tell me about the problem or the role.

Send a message

Start a conversation

For founders and startup teams: share the product stage and the decision you need help with. For recruiters: include the role, scope and team.

Online message forms are currently unavailable. Please contact me directly on LinkedIn.

Message Meet on LinkedIn (opens in a new tab)

Looking for my resume?

View my background or request a copy of my resume through LinkedIn.

Request my resume on LinkedIn (opens in a new tab)