AJAY VISHWANATHAN
Machine Learning Notes From Real Builds
Notes from building ML systems in production — the architecture decisions, the things that failed in ways benchmarks never predicted, and what actually held up.
Recent Writing
What I've been working through lately.
You Can't Improve What You Don't Measure: Evaluating Knowledge-Grounded Retrieval Systems
After shipping a RAG retrieval system, the hardest question is whether it actually works. This post covers RAGAS, LLM-as-judge calibration, the evaluation triad, and how to build an eval pipeline that gives you real signal.
Recommended Reads
A good place to start — posts that go deeper on things that actually mattered.
- You Can't Improve What You Don't Measure: Evaluating Knowledge-Grounded Retrieval Systems After shipping a RAG retrieval system, the hardest question is whether it actually works. This post covers RAGAS, LLM-as-judge calibration, the evaluation triad, and how to build an eval pipeline that gives you real signal.
- RAG in the Real World: Building a High-Precision Retrieval System for LLM Pipelines Building a high-precision retrieval system sounds simple on paper. This is what actually happens when you ship one — the chunking failures, the retrieval gaps, and the hybrid search that finally gave us reliable signal.
- The Pipeline Problem: Building the Data Foundation for an Agent That Doesn't Lie A story about graduating from ad-hoc ingestion scripts to a real distributed file-processing platform using Ray and Azure, and the hard lessons learned along the way.
Browse by Category
WRITING
Blog Posts
Long-form reflections on AI, software engineering, and the craft of building with models.
EXPERIMENTS
Practical Experiments
Hands-on notebooks and implementation notes from practical machine learning workflows.
RESEARCH
Paper Reviews
Structured breakdowns of influential research papers and their real-world implications.