About

I write software that other software depends on. Inference gateways, evaluation pipelines, real-time audio paths, ingest jobs. Basically architecting high throughput backends - the layer where a bad decision does not show up as a bug report, it shows up as a pager at 04:00 eighteen months later.

Right now I am a Senior Product Engineer (Staff Engineer) for AI and Backend at Raise Financial (Dhan) in Mumbai, leading the voice and video generative AI wing of Raise AI. Before that: data science and Go infrastructure at BlinkX by JM Financial, and explainable AI research with CSIR — National Chemical Laboratory, a Government of India lab, where the output was two papers, a software copyright, and a pending patent.

The short version of what I actually do: take a model that works in a notebook and make it survive contact with production traffic, a latency budget, and a compliance boundary. The full record is on the work page.

What I am GOATed at

Backend and data work in Python and Go, with Rust when the problem deserves it. GenAI systems end to end: retrieval pipelines, self-hosted inference on vLLM, model routing, and the evaluation harness that tells you whether any of it still works this week. Cloud on GCP and AWS.

The pattern across most of it is the same: the model is rarely the hard part. The hard part is the pipeline around it, the latency budget it has to fit in, and knowing when it has quietly stopped being correct.

Published work

Research from the CSIR-NCL collaboration, both in SCOPUS-indexed venues:

• Cancer XAI: A Responsible Model for Explaining Cancer Drug Prediction Models — IJISAE
• An explainable AI-assisted web application in cancer drug value prediction — MethodsX, Elsevier

Alongside those: one granted software copyright, one patent pending on the explainability methodology, and three more papers in the pipeline.

Credentials

Languages: English and Kannada natively, Hindi at full professional level, Telugu at limited working. Outside the terminal it is basketball, the gym, and other people’s tech blogs.

What this site is

Two things, deliberately kept in one place.

The work side is a record: where I have worked, what I built, what constraint made it interesting, what I would do differently. Short, and honest about the parts that did not go well.

The notes side is the useful half. Every entry starts as something that took me longer than it should have. If I had to read four issue threads and a mailing list post from 2011 to understand something, that is worth writing down properly once.

How it is built

Nothing exotic, and nothing off the shelf for the parts that matter.

The styling decision is the one I would defend hardest. The theme ships a compiled Tailwind bundle in @layer utilities; unlayered author styles beat layered ones regardless of specificity, so the whole design sits on top of it without a single !important. Every rule in the stylesheet is one somebody chose on purpose.

Anything you can describe in a design system, you can describe in about 900 lines of CSS. The difference is that you will know what all 900 do.

Elsewhere

Email is the one I actually read. Everything else is a mirror.

• GitHub  — source, mostly Go and Python
• LinkedIn  — the formal record
• shivanandanasharma@gmail.com — direct
• /feed.xml — RSS, still the best way to follow a blog

If you are reading this because something on the notes side was wrong: please tell me. Corrections are the highest-value mail I get.