genalpha

In one line

What it is.

A multi-asset portfolio for Indian markets, run by a team of GenAI agents that enforce both the discipline and the analytical depth retail investors most often fail at.

Why now

Why this is the GenAI moment for retail investing.

Retail investing fails on two compounding problems: behavior under stress, and information at scale. Both are textbook GenAI strengths — and only recently have models been good enough, cheap enough, and reliable enough to deploy this way.

• Behavioral discipline. Every position carries a written thesis with named invalidation. Agents enforce quarterly re-affirmation. Override patterns are logged and reviewed. Emotional drift becomes auditable instead of invisible.
• Information processing at scale. Annual reports, exchange filings, earnings calls, sector commentary, broker notes — parsed continuously. The "I didn't have time to read it" excuse disappears.
• Memory. Every decision, every rationale, every override lives in a versioned log. Patterns surface that humans working alone would miss.
• Continuous monitoring. Regulatory filings, exchange surveillance, credit-rating changes, insider transactions, auditor changes — tracked without weekends off.

Where GenAI moves the needle

What this enables that wasn't practical before.

Six concrete capabilities. Each one used to be "I'll do it later" or "I'll skim it." All of them now happen continuously, on every position.

Where the money lives

Asset allocation.

Mostly Indian equity. A bounded slice in US tech (the AI capex cycle is too structural to ignore). Bonds and cash for stability and dry powder. The universe is locked — no ad-hoc additions without a written amendment.

The math

Where the edge comes from.

Each piece is small. Together they aim for ~4% per year above a sensible mix. Picking actual Indian stocks is the dominant driver — if that breaks, nothing else compensates.

What ~4% extra actually means.

A reference: Indian large/mid-cap equities have historically delivered roughly 12-14% CAGR over long periods (Nifty 500 TR), with strong 5-year windows running ~18-20% and weak ones ~6-8%. An extra ~4% per year compounds far more than it sounds.

Numbers approximate; illustrative of compounding mechanics, not a guarantee of outcomes. Actual results depend on realized alpha, market regime, and execution.

The team

Six GenAI specialists. Two infrastructure agents. One human.

Each role has a clear job and a defined boundary. The owner approves every trade.

Risk discipline

What protects the downside.

A great strategy with bad risk management is worse than a mediocre one with good risk management. The rails matter more than the engine.

• −25% drawdown ceiling. A circuit breaker, not a hope. The protocol triggers at −15% — the system pauses new buys and routes contributions to bonds and cash, with daily drawdown attribution by sleeve.
• Every position has a thesis. Written reasoning for owning it, named conditions under which it would be wrong, expected return scenarios. No story without numbers.
• 5% hard cap on any single name. Force-trim if breached, regardless of conviction. Single-name blowups cannot compound.
• No autonomous trading. Agents recommend; the owner approves. Every trade. No exceptions for at least the first six months of operation.
• Quarterly mandatory review. Every position is re-affirmed in writing or trimmed/exited. Stale theses are auto-flagged. No drift, no auto-pilot.

The fun part

Hunting for the few things that go up a lot.

A small slice of the portfolio (~15%) goes into a basket of 12-18 small bets where each one could become a 2-5x in 2-3 years. Most won't. The few that do pay for the duds.

The edge is not in stock picking — most themes are visible to everyone by the time retail sees them. The edge is in being earlier than consensus, sizing right, holding through volatility, and exiting before the hype peaks. The agent harness is built to enforce all four.

Just so we're clear

What this is NOT.

A few things people sometimes assume an "AI portfolio" means. It doesn't.