#Synthetic Data — Kunal Ganglani

a man using a laptop computer on a table

How to Build Synthetic Data for RAG Evaluation

A leakage-first, adversarial-negative pipeline to generate, label, and validate a synthetic RAG evaluation set without building a benchmark that only tests your generator.

Keep exploring

Close to this topic, not on this page.

How to Test Non-Deterministic AI Systems and LLMs A release-gating playbook for LLMs and agents: golden sets, metamorphic tests, variance control, tolerance bands, cohort diffs, and… 10 min read
RAG Evaluation Metrics for Retrieval Quality: My Production Playbook If your RAG app got worse after an embeddings or chunking change, grading answers won’t tell you why. 11 min read
Agent Evaluation Roadmap for Small Teams: 30 Min a Week A pragmatic map of offline vs online vs HITL evals, what to measure beyond “response quality,” and the smallest program that actually… 8 min read