# Vishal Bhargava

> Platform engineering notes on running LLM systems in production — retrieval, evals, observability, and inference cost.

I help engineering teams get GPU and LLM inference costs under control through measurement, routing, and serving efficiency — without sacrificing reliability. 12 years in platform infrastructure. Based in Dubai.

This site is a personal engineering blog. It is written and maintained by one
person: Vishal Bhargava, a platform infrastructure engineer based in Dubai.
Posts are first-hand write-ups of systems actually built and measured, not
tutorials or aggregated summaries. Where numbers appear (latency, error rates,
cost), they come from the author's own instrumentation and are labelled with the
conditions they were measured under.

Every page below is also available as HTML at the same URL without the `.md`
extension.

## Writing

- [Hybrid search didn't help. It still fixed my retrieval.](https://vbhargava.org/writing/hybrid-search-ablation/index.md): I added hybrid search and a cross-encoder reranker to my RAG service, wrote up what improved, then ran the ablation I'd been avoiding. The pipeline had contributed nothing directly — but its 512-token limit forced a chunk-size change that did everything. (published 12 August 2026)
- [Building a RAG service: architecture, trade-offs, and lessons](https://vbhargava.org/writing/llm-lab-rag-v1/index.md): A production-shaped RAG v1: offline indexing + FastAPI /query on Cloud Run (Terraform), with sources, logs, and basic monitoring. (published 29 January 2026)

## About

- [About me](https://vbhargava.org/about/index.md): Background, current focus, and how to make contact.

## Optional

- [Homepage](https://vbhargava.org/): Short positioning statement and the most recent posts.
- [RSS feed](https://vbhargava.org/index.xml): Full-text feed of new writing.
- [ritam](https://github.com/bhvishal9/ritam): The open-source RAG/LLM-infrastructure project most of the writing draws on.