Vanshika Verma

At Acceleron Labs, I build RAG pipelines, LLM-driven agents, and the production infrastructure behind Vini Lite. I was promoted twice in 12 months from intern to Software Development Engineer I.

I designed Vini Lite's high-availability request-routing architecture with primary/backup DNS, active-active nginx load balancers, and stateless Flask nodes, sustaining 2,000 queries per second at the routing layer. I also built a DAG-based workflow engine over RabbitMQ quorum queues, with crash recovery and transactional deployment rollback for reliable stateful workflows.

I've improved retrieval accuracy by 10% through retrieval and ranking refinements, and built RAG search systems that reached 80% accuracy across 100 GB NFS/SMB file systems. For Vini, I built Milvus-backed retrieval and prompt improvements that achieved 90% answer accuracy across 200 evaluation questions with sub-three-second response times.

My work spans Python, Flask, React, Linux, LangChain, LangGraph, PostgreSQL, and distributed systems. I also developed RADAR, an ML-based IoT intrusion detection system that achieved 99.76% binary and 99.88% multiclass classification accuracy on CICIDS2017.