Apoorva Verma
Software Engineer · Applied AI
Software engineer building, measuring, and shipping AI systems end to end: retrieval, evaluation, and voice agents.
Who is Apoorva Verma?
Apoorva Verma is a software engineer who works on applied AI. She builds full-stack products end to end, then goes deep on the retrieval and evaluation work that keeps LLM systems correct in production, in Python and TypeScript. She's a software engineer at Oracle, after earlier stops at Microsoft Engage, J.P. Morgan, and Major League Hacking.
What does Apoorva Verma work on?
Four things. LLM evaluation and measurement. Retrieval and vector-search engineering. Real-time and voice AI agents built on LiveKit. And full-stack engineering for the products around them. The through-line is measuring what most people ship on vibes: reproducible benchmarks, calibrated judges, and answers that cite their sources.
What has Apoorva Verma built?
Her RAG evaluation harness (rag-eval-harness) introduced a citation-faithfulness metric that never touches the gold set. Framed the way the ALCE attribution-eval literature does it, it audits an answer against its own cited sources instead of a reference answer. The same project documents her retrieval ladder, dense to BM25 hybrid to cross-encoder rerank, which lifted NDCG@5 from 0.33 to 0.62 across 50 questions. She also ships packages on PyPI (langchain-shannonbase and llama-index-vector-stores-shannonbase) and a LiveKit plugin for voice agents.
