- RAG and knowledge assistants
- Search and ranking
RAG retrieval quality benchmark
Answer correctness 0.804 → 0.913 (baseline → advanced)
Three retrieval setups, one corpus, 55 test questions. Hybrid search and reranking gave the clearest gain.
AI now reads a company's data and helps build it, and how well it works depends on the data models, definitions and tests underneath. Innervu builds both: the AI and the foundation it needs.
You work directly with And, from the first call to handoff, with weekly updates written in plain language.
How we workEach project reports its results against a baseline, with the sample size and what it can't tell you.
Answer correctness 0.804 → 0.913 (baseline → advanced)
Three retrieval setups, one corpus, 55 test questions. Hybrid search and reranking gave the clearest gain.
A Claude Skill wrapped in a small web app. Users upload a file, click Run and download the result without installing the skill.
Free, 30 minutes
We talk through the problem and the goal, and check fit honestly.
Paid, fixed price
We map the process and the data, measure the baseline and agree on success metrics.
Fixed scope
Working software early, a short demo every week and a test set built with you.
We go live, train the people who use it and document how it works.
Optional, monthly
Monitoring and improvements, with the test set re-run when models or data change.
Founder, Innervu
And runs Innervu and works with every client directly, from the first call to handoff.
Before Innervu, And was AI/ML Director at Quizlet, leading the recommendations and personalization machine learning team, and Head of Recommendations Platform and Science at Etsy.
And teaches AI Workflows for Non-Programmers (ETM 472/572) at Portland State University.