I'm an AI Engineer specializing in building production-grade agentic AI systems, multi-agent orchestration pipelines, and RAG-powered applications. My work sits at the intersection of LLM engineering, prompt design, and scalable backend architecture — turning cutting-edge research into systems that actually work.

• 🤖 Building autonomous agent loops with LangGraph, LangChain, and CrewAI
• 🔍 Designing RAG pipelines with evaluation frameworks (RAGAS, Evals) and vector stores (ChromaDB, FAISS)
• ⚡ Deploying AI backends with FastAPI and serving UIs via Gradio, Streamlit, and HuggingFace Spaces
• 🧬 Working with Gemini, GPT-4o, NVIDIA NIM, and HuggingFace models
• 📫 Reach me at vishwa0r1@gmail.com

Autonomous agentic platform powered by the DS-STAR Framework (Self-Taught Agent with Reasoning)

An end-to-end multi-agent system that autonomously analyzes datasets, generates Python code, executes it in a Docker sandbox, and self-verifies output against user constraints — all in a real-time streaming loop.

• Agent Pipeline: File Analyzer → Planner → Coder → Executor → Debugger → Verifier → Router → Finalizer
• Secure Sandboxing via isolated Docker containers for AI-generated code execution
• Live Streaming of agent states and artifacts via Server-Sent Events (SSE)
• Sub-question decomposition for complex multi-query workflows
• Python FastAPI LangChain NVIDIA NIM React 19 Supabase Docker

Hierarchical multi-agent system for unified customer support using LangGraph StateGraph

A production-style multi-agent architecture with a Super Agent orchestrating 4 domain-specialized agents (Orders, Logistics, Payments, Support) — each powered by Text-to-SQL over isolated virtual databases.

• 40% latency reduction via intent caching; 60% of queries bypass LLM with pattern matching
• Parallel agent execution using ThreadPoolExecutor
• Conversation memory via MemorySaver checkpointing
• Real-time web chat UI
• Python LangGraph Django GPT-4o SQLite Multi-Agent Systems

AI-powered code review system using a CrewAI 5-agent pipeline with Google OAuth

Full-stack application that automatically reviews GitHub pull requests by dispatching specialized AI agents — Diff Analyst, Code Reviewer, Security Analyst, Best Practices Advisor, and Technical Lead — each independently analyzing the PR and synthesizing a final verdict.

• 5 specialized CrewAI agents with role-based prompt engineering
• Google OAuth 2.0 authentication with JWT-secured API
• Fallback pattern-based analysis when LLM is unavailable
• Python FastAPI CrewAI LangChain Next.js 15 TypeScript Firestore