Rohit Singh

I build AI systems that don't just work, they think, recover, and improve on their own. Over 8 years I've architected production-grade agentic AI systems across fintech, insurtech, retail, e-commerce, accounting, and travel, systems that handle real workloads, real edge cases, and real business consequences. Not demos. Not prototypes. Production. What that looks like in practice: A multimodal pipeline that processes handwritten financial documents, cut processing time from 4 hours to 8 minutes using Claude Opus, AWS Nova, and a custom CNN for page classification A production support agentic system with contextual event memory that autonomously resolves 70.2% of tickets, routing the rest to human-in-the-loop with full context. An intelligent document processor with dual-LLM architecture and LLM-as-Judge validation achieving 94.7% accuracy across invoices, bank statements, and regulatory documents, reducing manual processing time by 78.3% A self-cure agentic system for autonomous incident detection, root cause analysis, and automated remediation, four-phase architecture: Code Indexing → Retrieval → Reasoning → Remediation Beyond building systems, I go below the API surface. Currently finetuning Gemma 4 and Qwen 3 8B for domain-specific document understanding, because off-the-shelf models aren't always enough when production accuracy is non-negotiable. I also work at the frontier of AI-assisted engineering, using Claude Code, Cursor, and AWS Kiro daily to architect and ship faster without cutting corners on quality. This isn't a tool preference, it's how senior AI engineering gets done in 2026. How I build: LangGraph · LangChain · LlamaIndex · MCP · AWS Bedrock · SageMaker · Agentic Core · Claude · OpenAI · Gemini · Llama · MilvusDB · ChromaDB · Langfuse · FastAPI · Python What I'm open to: Senior and leadership roles (remote, EU