.png&w=256&q=75)
RAJEEV KUSHWAHA@rajeev10200300093483
5
Events attended
2
Submissions made
India
1 year of experience
About me
I’m Rajeev, a Computer Science student passionate about **AI, Machine Learning, and technology**. I enjoy building real-world projects, exploring new technologies, and continuously improving my problem-solving and development skills. I’m also actively involved in creative and social media initiatives, where I love combining technology with creativity.
🤓 Latest Submissions
.png&w=828&q=75)
InvestigateOS: Voice Investigation Simulator
InvestigateOS is a voice-driven investigation simulator designed for situations where understanding what happened requires more than asking an AI a question. Users enter a simulated ML production incident and investigate it naturally through voice. AssemblyAI provides the real-time voice interaction, while an investigation agent retrieves evidence, reasons over timelines, builds hypotheses, searches for counter-evidence, and explains its conclusions. Unlike a traditional voice chatbot or RAG system, InvestigateOS maintains a deterministic World State and Evidence Graph. The AI does not directly control the truth. It investigates evidence and proposes actions, while the simulation engine validates and executes state changes. For example, when an ML model's F1 score drops, the user can ask what changed before the degradation, inspect the evidence linking a feature-pipeline change to the incident, challenge the hypothesis with counter-evidence, and then say “Rollback the feature pipeline.” The simulated system changes state, generating new events and evidence that can be investigated again. The result is a closed-loop experience: Voice → Investigation → Evidence → Hypothesis → Action → State Change → New Evidence.
30 Sep 2026
Autonomous Alpaca Options Trading Desk
**Autonomous Alpaca Options Alpha Agent** is an autonomous AI-powered options trading system built for the Alpaca AI Trading Agents Hackathon. The project combines LLM-based market reasoning with deterministic risk controls to create a safer and more auditable trading-agent architecture. The agent continuously retrieves live market and options data from Alpaca and scans potential opportunities based on liquidity, volatility, momentum, and options quality. Selected opportunities are passed to an OpenRouter-powered AI strategy engine, which evaluates strategies including Long Calls, Long Puts, Bull Call Spreads, and Bear Put Spreads. Rather than allowing the LLM to directly execute trades, its recommendations are converted into structured JSON and validated using Zod. Every proposal then passes through deterministic risk gates covering AI confidence, maximum position risk, portfolio exposure, buying power, active position limits, and daily loss protection. Trades that fail any safety condition are automatically rejected. The system runs through an autonomous agent loop that handles market scanning, AI reasoning, validation, risk evaluation, execution, position monitoring, and decision logging. Alpaca MCP and CLI fallback support the integration and execution layer, while unique client order IDs help prevent duplicate orders. A React-based dashboard provides real-time visibility into agent status, opportunities, positions, risk decisions, and trading activity. SQLite maintains a persistent audit trail of market data, AI decisions, risk checks, and executions. **Tech Stack:** TypeScript, Node.js/Express, React, Vite, Tailwind CSS, SQLite, Alpaca Trading API, Alpaca MCP, OpenRouter, Docker, and GCP Cloud Run. The system is built and tested entirely in **Alpaca Paper Trading** with no real capital involved. It is an experimental hackathon project and not financial advice.
4 Sep 2026
