Financial Markets / Proprietary Research
Proprietary AI Trading Desk
Our internal trading terminal. 8 autonomous agents, 15 markdown rule files, zero algorithmic code. A personal hedge fund desk running on Indian markets. Research and proprietary use only.
What This Is
NiftyX is our proprietary trading desk. Not a product. Not a service. A personal terminal — built for one operator, running live on Indian markets every day the market is open.
The goal was simple: build the infrastructure of an institutional trading desk without the institution. One engineer, an AI team, and a disciplined rules framework — operating at a level that previously required a quant team.
What Made It Hard
No algorithmic code. Traditional bots use hard-coded rules that break when market regimes change. NiftyX uses LLMs to reason through 15 markdown rule files — each encoding a specific trading thesis in plain English. The LLM reads market data, reads the rules, and surfaces analysis. Strategy updates are English edits, not code deploys.
8 agents coordinating in real-time. Scanner watches 200+ instruments. Analyzer deep-dives candidates. Risk Manager validates position sizing. Trade Executor fires on Kite. Position Monitor tracks open positions. Alert System pushes to Telegram. Journaling and Review agents close the loop. Each has its own context, tools, and failure modes.
SEBI compliance baked in. Automated trading in India has strict guardrails — position limits, order frequency caps, market-hour restrictions. These aren't afterthoughts; they're in the agent governance layer.
The Architecture
8 agents: Scanner, Analyzer, Risk Manager, Trade Executor, Position Monitor, Alert System, Journaling Agent, Review Agent. Each runs as an independent Claude API call with specialized tools.
Data: Kite Connect for real-time market data and order execution. Supabase for trade journal, agent logs, and position tracking. 15 markdown rule files as the strategy brain.
Infrastructure: Dedicated VPS. Telegram bot for alerts and manual overrides. Runs autonomously during market hours with no human in the loop.
The Outcome
Live daily since March 2026. 106 commits. Identifies, analyzes, and executes without intervention. Every trade journaled with full agent reasoning chains. Strategy evolves through plain English rule edits — no redeployment needed.
This is what it looks like to run a hedge fund desk as a solo operator.
Tech Stack