Anant Bansal

I've built AI-native products from core RAG pipelines to production-ready memory layers. At a stealth startup, I worked directly with a five-person founding team to ship three major iterations of an AI productivity assistant in 12 weeks.

I fine-tuned Llama 3 and Mistral 7B for task automation, improving intent classification by 40% while cutting inference cost by 35% through 4-bit quantization and vLLM. I also built a Qdrant and Redis semantic cache and deployed an autoscaling inference stack on AWS EC2.

At Microsoft, I engineered target-based autoscaling for Azure Functions using real-time Prometheus metrics and built a C# LLM PR-review agent that reduced review turnaround from four hours to under 90 minutes across 15+ internal repositories.

I'm drawn to end-to-end ownership and high-velocity product development. My projects include Luminous, an open-source context and memory layer for AI apps, and CryptX, a crypto monitoring platform that acquired 30+ paying users in its first three weeks.