Full-Stack TypeScript Engineer | Real-Time Systems & Agentic Workflows

I build high-performance web applications and contribute to the infrastructure that powers them. I specialize in TypeScript, WebSockets for low-latency streaming, and orchestrating LLM tool-calling pipelines.

• 🔭 Currently building complex AI agent architectures and real-time backend systems.
• 🌱 Deep-diving into Rust-based actor models and optimizing open-source LLM orchestration frameworks.
• 📫 Connect with me on LinkedIn or reach out at amritraj4work@gmail.com.

1. Stock Trading Platform A low-latency trading engine handling concurrent orders and live tick data.

• Architecture: Built with TypeScript and Node.js. Designed a multiplexed WebSocket streaming architecture to push real-time market data to the React client without overwhelming the main thread.
• Why it matters: Demonstrates strict concurrency management, preventing race conditions in financial state management.

2. Multi-Workspace Document Assistant (RAG) An enterprise-grade retrieval-augmented generation (RAG) pipeline with autonomous tool calling.

• Architecture: Implemented complex data chunking, vector similarity search, and a robust prompt-chaining loop to prevent LLM hallucinations.
• Why it matters: Proves ability to move beyond basic API wrappers into robust AI system orchestration.

3. Universe Explorer 3D An interactive WebGL simulation mapping realistic orbital movements.

• Architecture: Built using React Three.js. Implemented strict memory management within the browser's animation loop (requestAnimationFrame) to prevent memory leaks during complex camera matrix transformations.

4. AI Medical Consultation Voice LLM A voice-to-voice AI co-pilot for medical consultations.

• Architecture: Engineered a real-time asynchronous pipeline handling audio streaming, live transcription, and text-to-speech with stringent low-latency constraints.

I actively contribute to distributed orchestration and agentic AI frameworks. I focus on architectural improvements, LLM reliability, and performance bottlenecks.

• aden-hive/hive: Engineered an LLM self-correction loop within the orchestrator, automatically feeding validation errors back to the model to improve tool-calling reliability. View Pull Request
• rivet-dev/actors: Engineered external OIDC/JWT validation inside onAuth lifecycle hooks to support enterprise authentication. View Pull Request
• rivet-dev/actors: Identified and benchmarked a severe O(N) array lookup CPU bottleneck occurring under high load in the runner tunnel. View Architecture Issue