Plouton Strategy

A note on how we've presented this. Some of this work was done inside firms our team members worked at. Some was live client work. Some was independent. Each case says which. We don't take credit for a whole firm's outcome, and we don't put a client's name on this page without asking them first.

01
Venture diligence · Sparrow Capital · Krrish Garg

Sizing a market from the bottom up, and closing a ₹3.3 crore round

The problem

An early-stage air conditioning hardware company needed a valuation and a market size an investor would actually believe rather than politely nod at.

What we did

Sized India's ₹18,000 crore consumer air conditioning market from the bottom up rather than taking a percentage off a headline figure, arriving at a ₹750 crore serviceable market. Built a working pro-forma model to pressure-test the unit economics. Co-wrote a 15-page investment memo covering the global market dynamics that would affect the business.

What came of it

4.75x LTV to CAC and 18 months of runway, both stress-tested rather than asserted. The round closed at ₹3.3 crore.

02
Quantitative research · Tusk Investments · Krrish Garg

A trading strategy that beat buy-and-hold by a wide margin

The problem

Vedanta's share price moves with commodity prices. Everyone on the desk knew that. Nobody had a clean way to measure how much, or a way to trade on it.

What we did

Built VEDIX, an index of eight commodities weighted by how much each contributes to Vedanta's EBITDA, across 968 trading days from 2022 to 2026. Regressed the share price against it. Tested seven forecasting models and kept only the one that held up on data it hadn't seen. Then wrapped that model in a risk engine that reads market regime and volatility before deciding position size.

What came of it

The index explained 82% of Vedanta's price movement. The chosen model reached 97% out-of-sample accuracy and called direction correctly 60% of the time. The full strategy produced a Sharpe ratio of 4.02 against 2.40 for simply holding the stock, and cut the worst drawdown by 69%, down to 4.9%.

03
Portfolio strategy · MicroSave Consulting (India and Singapore) · Krrish Garg, Devansh Agarwal

Turning $1 into $2.20 across ten years of live market data

The problem

A $1M mandate across seven assets. The client wanted allocations that responded to market conditions, rather than a fixed split rebalanced once a year and hoped over.

What we did

Built a market-regime detection model — a four-state Gaussian hidden Markov model reading four macro indicators over a ten-year window. Checked it against six historical crashes to confirm it flagged them early rather than after the fact. Backtested four allocation strategies monthly from 2015 onward. Used a 252-day rolling window so no part of the test could accidentally use information from the future.

What came of it

The best strategy grew $1 into $2.20. Regime-aware weighting delivered 43.05% through the bull run and a 29.51% recovery bounce, ahead of four fixed-weight alternatives. The client also received a benchmark of five wealth platforms managing between $10B and $286B, with fee structures from 0.10% to 1.25% and returns on equity up to 28.3%.

04
Venture diligence · Sparrow Capital · Devansh Agarwal

Catching a 10x mispricing before the cheque was written

The problem

A seed-stage B2B fintech was being valued off a top-down market number. Those numbers are almost always too generous, and nobody had checked this one.

What we did

Sized the $38.9 billion B2B payments market down to a $3.07 billion serviceable market using SME share, digital adoption rates, and marketplace penetration. Then rebuilt the same number from the bottom up, starting from unit economics.

What came of it

The two numbers didn't match. Reconciling them exposed a ₹324 crore error in total payment volume and a target priced roughly 10x above what the underlying business supported. We recommended a valuation below 5.7x EV/Revenue, anchored on PayMate as the closest comparable, with a 3-7x band tested for the downside.

05
AI engineering · Design Industries, Australia · Nidhi Bharti, Mayank Goel

Automation that writes its own automations

The problem

A consulting firm was rebuilding the same internal scaffolding by hand every time it needed a new tool. It also had no reliable way to check whether the AI systems it shipped were behaving, or whether the ones it shipped last quarter still were.

What we did

Built retrieval workflows that treat the firm's Jira and Confluence as live memory, so tools answer from what the company currently knows rather than a stale export. Built a pipeline that reads a written specification in Jira and produces the working automation from it. Added a governance layer that audits existing automations, rewrites the ones that have drifted from spec, and republishes them. Put a four-layer safety check in front of model output and documented what it caught and why.

What came of it

Invalid model responses down 30%. Document extraction errors down 35% once schema validation was in place. 500+ unstructured PDFs processed through a pipeline with no manual handling and no credential exposure. The safety layer caught 87% of unsafe prompts across 100+ test cases covering prompt injection, personal data leakage, abuse, and jailbreak attempts.