I’m an AI Software Engineer passionate about bridging the gap between research and production. I specialize in building low-latency, scalable Retrieval-Augmented Generation (RAG) systems and Real-Time Voice AI applications. My work focuses on designing robust architectures that integrate LLMs, vector search, and agentic workflows to solve real-world problems.

• Retrieval-Augmented Generation (RAG): Designing systems with robust retrieval, cross-encoder re-ranking, and hallucination control.
• Agentic AI Workflows: Building reasoning engines using LangGraph and tool-driven architectures.
• Real-Time Voice AI: Integrating Whisper ASR, Silero VAD, and streaming TTS for seamless conversational experiences.
• End-to-End AI Pipelines: Taking projects from rapid prototyping to production deployment on AWS.

• Languages: Python, TypeScript, JavaScript, SQL
• AI & ML Frameworks: LangGraph, OpenAI, Whisper, Silero VAD, MediaPipe, OpenCV
• Backend & Cloud: FastAPI, Flask, Node.js, AWS, WebSockets
• Databases & Vector Search: Qdrant, FAISS, MySQL

Built a multi-LLM RAG system focused on accuracy and relevance.

• Implemented cross-encoder re-ranking and optimized chunking strategies to reduce hallucinations.
• Integrated Qdrant/FAISS for efficient vector retrieval.
• Supported multimodal extensions, including voice interfaces.

Designed a low-latency, bi-directional audio streaming architecture.

• Orchestrated WebSockets for continuous audio streaming.
• Integrated a full speech pipeline: Silero VAD -> Whisper ASR -> LLM Reasoning -> Streaming TTS.
• Optimized for minimal latency and natural turn-taking.

Blending cultural art with modern computer vision.

• Developed a real-time recognition system using OpenCV and MediaPipe.
• Achieved ~90% accuracy in identifying classical dance hand gestures.
• Built an interactive Flask application for live inference visualization.

• Model Context Protocol (MCP): Investigating advanced patterns for standardized context exchange.
• Multi-Agent Systems: Designing collaborative agent architectures for complex task solving.
• Scalable Real-Time Architectures: Optimizing WebSocket handling for high-concurrency voice AI.

I built a real-time Bharatanatyam gesture recognition system to combine my interest in computer vision with traditional Indian art forms!