At Think Future Technologies, I built modular ETL pipelines that bring Gong, CRM, Zendesk, and other customer data into a shared account timeline. I also developed LangGraph workflows to extract risks, sentiment, commitments, and expansion opportunities from customer calls and tickets.
I built GenBench, a pip-installable Python SDK for repeatable generative AI benchmarks across OpenAI and Vertex AI models, with configurable metrics including LLM-as-a-judge evaluation.
At Genzeon, I migrated inference to vLLM across multiple NVIDIA RTX 5090 GPUs, cutting per-file inference time by 60% while scaling to 12,000+ pages per day. I also designed hybrid search and extraction pipelines that raised average extraction accuracy from 62% to 87%.
Earlier at Antier Solutions, I built document-processing and semantic search systems, including a RAG chatbot over product catalogs. My work has spanned NLP, information retrieval, and production AI services using Python.
