Looking for a job

Shivam Rai

@shivamrai2

Message

AI/ML Engineer at Oolka building hybrid RAG systems and benchmarking memory retrieval.

India

Message

At Oolka, I built a memory-retrieval benchmark and compared embedding models on retrieval quality, latency, and storage. I used an abstention margin to identify a model’s limits with missing memories, then selected a self-hosted Qwen3-0.6B setup with reranking.

I also built a hybrid RAG system and a BI tool for mining chat history. Previously, at HelloAgentic AI, I architected a multi-agent pipeline and built a design agent that converted screenshots into production-ready code.

Experience

Work history, roles, and key accomplishments

Current

AI/ML Engineer

Current

Oolka

May 2026 - Present (5 months)

Built a memory-retrieval benchmark and evaluated embedding models, achieving high nDCG@10 with self-hosted Qwen3-0.6B. Profiled CPU-only LLM inference and load-tested embedding servers, contributing to an event-driven task pipeline and hybrid RAG system.

Python Next.js EC2 PM2 LLamacpp RAG Embeddings Reranking Retrieval Evaluation NDCG@K Recall LLM Inference Optimization Event Driven Architecture

Software Developer Intern

HelloAgentic AI

Dec 2025 - Mar 2026 (3 months)

Architected a multi-agent pipeline with FastAPI and async workers, built a real-time collaborative editor, and engineered a Super Agent routing queries to reduce token costs by 30%.

fastAPI WebSockets Multi Agent Systems GPT 4o Claude LLaMA Cascading Classification Token Usage Optimization