// Personal project
Project Kuber
I built a multi-agent trading intelligence system for the Indian stock market, delivered entirely through a Telegram bot instead of a website.
Tech Stack
How it works
Multiple specialized agents independently analyze technicals, fundamentals, news sentiment, and overall market direction, then combine their outputs using a fixed, deterministic rule set rather than AI guesswork.
- Ran all financial calculations directly in Python so the numbers are always verifiably accurate, with AI used only to explain results in plain English.
- Built automated scheduled reports: a morning briefing, hourly checks during market hours, and an end-of-day summary.
- Built a running record of the bot's own past calls to track prediction accuracy over time.
Why I built it
AI trading tools often let the model do the actual math, which makes the numbers hard to trust. I wanted a system where the financial calculations are always deterministic and verifiable, and AI only handles the explanation, not the arithmetic.
How I approached it
I split analysis across specialized agents for technicals, fundamentals, sentiment, and market direction, combined their outputs with a fixed rule set instead of letting an AI vote on the outcome, and delivered everything through Telegram instead of a separate website.
