---
title: Production ML systems expertise — Sudhanva Narayana
description: Evidence-backed expertise in ML platforms, inference, serving, and reliability.
canonical: https://sudhanva.me/expertise/production-ml-systems/
last-updated: 2026-10-01
---

# Production ML systems expertise — Sudhanva Narayana

Sudhanva Narayana designs and operates production ML systems that move models
from experimentation into observable, repeatable workloads. His public evidence
spans billion-row batch inference, real-time transformer serving, multi-GPU
delivery, distributed fine-tuning with PyTorch DDP, Kubernetes-based platform
engineering, and multi-regional geospatial ML.

## Capabilities

- Production ML platforms with Kubernetes, Ray, Flyte, Terraform, Argo CD,
  experiment tracking, and model versioning.
- Batch and real-time inference, multi-GPU execution, model serving, input
  pipelines, performance tuning, and deployment automation.
- Distributed training with PyTorch DDP, torchrun, and NCCL
  collectives, LoRA and PEFT fine-tuning of LLMs such as Gemma 4, GPU memory
  budgets, and choosing between data, tensor, pipeline, and FSDP parallelism.
- Reliability practices including monitoring, alerting, recovery, safe
  rollouts, data validation, resource controls, and cloud-cost management.
- Scientific and data-intensive ML, including vector search and geospatial
  analytics.

## Evidence

- More than 10B rows processed by a TensorFlow prediction workflow in under
  three hours.
- More than $500K in annual cloud-cost reduction for production ML
  infrastructure.
- Twice the model release velocity, an automated build, tuning, and deploy
  pipeline in place of recurring manual work, and lower GPU cost per release through an orchestrated multi-GPU Kubernetes
  delivery platform with CI/CD.
- More than 1M daily interactions analyzed by a real-time transformer pipeline
  ([write-up](https://sudhanva.me/blog/posts/real-time-transformer-prediction-at-autodesk/)).
- A PyTorch DDP training script for LoRA fine-tuning Gemma 4 12B across GPUs
  ([engineering study](https://sudhanva.me/work/case-studies/pytorch-ddp-gemma-4-fine-tuning/)).

- [Complete expertise page](https://sudhanva.me/expertise/production-ml-systems/)
- [Production case studies](https://sudhanva.me/work/)
- [Technical writing](https://sudhanva.me/blog/)