A low-cost proprietary trading desk built for capital preservation — throwing off cashflow that funds a gated bet on AI-run trading.
This is not a fund raise and claims no track record. The edge sold is tight, centralised risk management and institutional-low transaction costs via direct exchange membership — not an unproven house strategy. The desk is engineered to make money on its own; the AI programme is optionality layered on top, funded from real cashflow rather than raised against a promise.
How the model works
Structured as an LLP with direct exchange membership, so costs are institutional-low. Five employed traders each receive a ₹10 Cr capital limit — ₹50 Cr in total — and run their own short-term index books. Funding is ₹40 Cr of arm's-length investor capital at a fixed 9%, parked in G-Sec (6.7%) and pledged for exchange margin; a 20% cross-margin benefit cuts the net margin required to ₹40 Cr. Traders post a 10% deposit (₹5 Cr) as first-loss capital, ahead of ₹50 lakh of operator equity.
Because the parked G-Sec already earns most of the coupon, traders need only ~4.24% gross — net of transaction costs — to keep the desk alive. In the base case the desk clears ~₹2.54 Cr profit after tax and retains ~₹2.03 Cr after funding AI R&D.
The AI bet — training and transition
The desk is a cashflow engine that funds a parallel, gated R&D programme: 20% of each year's PAT (~₹51 lakh at base) builds an AI multi-agent trading system that coordinates like an institutional floor. The desk supplies what money can't easily buy and models most need — real fills, real risk events, real P&L to train and validate against.
The transition is staged and evidence-gated: agents earn allocation only as they clear a dated validation roadmap — first shadowing books, then running risk-capped sleeves, then scaling if and only if they beat the human benchmark net of costs. No capital is ever bet on a promise.
Tail events — both directions
Downside. Books are ~90% Nifty/Sensex, so in a gap they are correlated and blow up together rather than diversifying — the buffer is therefore sized to the aggregate, not per trader. RMS enforces a hard aggregate stop (~6.75%) and the desk carries OTM index puts as a standing tail hedge. Modelled stress (−3%, after stops and hedge) is absorbed by the ₹5.5 Cr first-loss buffer with ~₹1.88 Cr to spare. The buffer is deliberately sized hedge-free; the disclosed worst case is a naked overnight gap that skips the stop toward −10%, where the buffer takes a dent of ~₹1.62 Cr (≈4% of investor principal) — a residual we either fund with larger deposits or disclose, never hide.
Upside. The same convexity that hedges the tail pays asymmetrically in a dislocation, and the AI programme is itself a cheap call option — a small slice of PAT with a potentially outsized, compounding payoff. Boring floor, convex top.
Even if AI fails, this is a boring business that makes money
Strip out the AI entirely and what remains is a profitable, low-cost prop desk: institutional transaction costs, a first-loss structure that protects investor capital, and a breakeven trader return of ~4.24%. Capital preservation is the product; the AI is an asymmetric bet the business can afford to make because the floor pays for it. The downside of the bet is capped at the R&D allocation; the upside is uncapped.
A lever: equity in place of some G-Sec for alpha
Today 100% of investor capital sits in G-Sec at 6.7% — safe, liquid, cleanly pledgeable for margin. A deliberate variant is to rotate a measured slice into large-cap equity to lift the blended carry and add alpha above the coupon. The trade-off is explicit: equity carries higher margin haircuts, more mark-to-market volatility, and correlation with the very index the desk trades — so any allocation is sized against margin cover and the first-loss buffer, not bolted on for yield. A lever to pull with intent, not a default.
Where it stands
Investor capital is arm's-length debt. Whether investors are best structured as LLP partners or lenders is a live legal question — a fixed return on pooled trading capital may be a regulated activity — and sits with counsel. The purpose of this plan and dashboard is credibility and a dated validation roadmap: show the risk architecture, let a sceptic stress it live, and prove the floor before scaling the bet.
A proprietary trading desk that sells tight risk management and low-cost exchange access, not an unproven house strategy. Investor capital sits in G-Sec, earns interest, and is shielded by a first-loss buffer. The one question this page answers: at what point does a bad year reach investor principal — and does the buffer get there first?
First-loss waterfall
Losses are absorbed bottom-up: trader deposits first, then operator equity, then — only if both are exhausted — investor capital. The striped band is the modelled loss for the active scenario. The headroom between the loss line and investor capital is the product.
The survival spread
The desk borrows at the investor rate and earns from trading plus the G-Sec its collateral sits in. So the trading return needed to survive is lower than the coupon you pay — the naive "beat the coupon" spread misreads it.
Stress it yourself
Drag any input — every number updates live. Slide trader return until PAT reaches zero to find the breakeven point for the current coupon, deposit and costs.
Where the money sits — and what backs it
Margin cover
Investor money is held in G-Sec and pledged for exchange margin. Cover is collateral (net of haircut) plus cash, against the margin required after the cross-margin benefit.
Proven traders, a hurdle they must beat, and your money kept behind a buffer
The desk does not bet on an untested house strategy. Traders get a capital limit, must first beat a hurdle, and share only the profit above it. Investor money sits in safe G-Sec, earns interest, and is protected by a first-loss buffer before it is ever at risk.
Your position as an investor
Four return worlds, one buffer
First-loss waterfall & risk limits
Clean-exit protection vs gap risk
A hard aggregate stop protects investor capital only if the market lets you exit at it. The gap between the stop-protected level and the modelled stress is the residual tail you either fund or disclose.