Abhinav Singh

I'm a full-stack AI/ML engineer who builds production LLM, RAG, and agentic systems end-to-end, from backend services to frontend. I currently work on a customer-facing agentic chatbot on AWS Bedrock AgentCore, targeting 150,000+ monthly users, and have shipped two other AI platforms serving 5,000+ daily users across 3 countries.

On DanskeAssist, a RAG knowledge assistant, I architected a multi-region Elasticsearch setup with kNN semantic search and per-tenant index isolation, serving multiple regions from a single API. I reduced query resolution from 6 minutes to under 1 second for 5,000+ daily users, validating the system against 611 evaluated Q&A pairs before launch, and later led a zero-downtime embedding migration from ada-002 to text-embedding-3-large after measuring retrieval quality degradation on multilingual queries.

On GRASP, a multi-agent compliance platform, I built a 27-endpoint FastAPI backend orchestrating 6 specialized agents through a five-phase conversation state machine, with parallel function calling and DynamoDB session persistence. As sole developer, I took it from scratch to production, cutting a 3-month manual process down to 15-20 minutes for 400+ users.

On the AgentCore chatbot, I built the full ingestion pipeline (S3, Bedrock Knowledge Base, OpenSearch Serverless) with Terraform and GitHub Actions CI/CD, cut response latency from 6-8 seconds toward a sub-3-second target, and built an LLM evaluation framework on Phoenix and OpenTelemetry to track quality, latency, and cost. My core stack is Python, FastAPI, LangChain/LangGraph, AWS Bedrock, OpenSearch, DynamoDB, and Terraform.