At Trajector, I revamped the AI document information extraction pipeline, reducing LLM inference costs by 60% and processing time by 80%. I also built an evaluation harness to benchmark extraction accuracy against ground truth.
I designed a call-info extraction pipeline that replaced manual data entry, saving 60–70% of the time previously spent on it. I built an AI Draft Creator service for legal and medical documents, using multi-step prompt pipelines across 54 domain templates to reduce average draft creation time by 70%.
Earlier, at UOLO.com, I worked on backend development for Tekie, its flagship ed-tech platform, and built systems including a modular onboarding flow and hybrid access control. My projects include pretraining GPT-2 (124M) from scratch and engineering an asynchronous multi-agent LLM orchestration framework.
