Atulya Kandpal@atulya_kandpal367
3
Events attended
1
Submissions made
India
1 year of experience
About me
Hi, I’m Atul Kandpal, a college student with a strong interest in software development, artificial intelligence, and emerging technologies. I’m currently building my programming fundamentals with C/C++ and Python while expanding my skills in AI/ML, web development, and Git/GitHub. I enjoy working on practical projects and hackathons where I can turn ideas into working solutions. I’ve worked on projects involving AI-based applications, machine learning models, and software development, and I’m particularly interested in using AI to solve real-world problems. I’m continuously improving my understanding of programming, problem-solving, version control, and AI technologies, while gaining hands-on experience through projects and collaborative development. My goal is to become a strong software engineer capable of building useful, scalable, and intelligent applications.
🤓 Latest Submissions
TradePilot: Explainable AI Trading Agents
TradePilot is an accountable multi-agent AI trading system that separates intelligence from execution authority. A Gemini-powered Strategy Agent analyzes structured market data and produces trading proposals with BUY, SELL, or NO_TRADE decisions, confidence, thesis, evidence, assumptions, validation conditions, and a suggested stop. Instead of allowing an LLM to directly control a broker, TradePilot passes every proposal through a deterministic Risk Engine that evaluates exposure, buying power, position constraints, and execution limits. The AI cannot override these safeguards. Approved trades then require explicit human confirmation before being sent to Alpaca paper trading, with live trading disabled and a hard $100 demo execution cap. TradePilot records the complete trading lifecycle through an explainable Decision Trace, connecting market evidence, AI reasoning, risk authorization, human approval, Alpaca order IDs, execution status, and actual broker fill data. This creates a verifiable audit trail showing not only what the AI proposed, but why the trade was allowed and what actually happened. The system also includes a post-trade Forensics and TradeMemory loop. Once a trade closes and a real outcome is available, the Forensics Agent analyzes the result and produces structured lessons that can be stored and retrieved as context for future decisions. TradePilot does not fabricate P&L or lessons from open or hypothetical trades. The system has been demonstrated with a real PLTR Alpaca paper trade, where the AI proposed BUY with 75% confidence, the Risk Engine approved it, human confirmation was provided, and Alpaca filled 0.5369 shares at $184.61. TradePilot: AI proposes. Risk controls. Human approves. Alpaca executes. The system learns from reality.
4 Sep 2026
