Hey! I’m Shubham — a Software Engineer who builds the backend for your backends. APIs, infrastructure, the stuff that sits behind the stuff users actually see. I’ve spent the last few years deep in MLOps and platform engineering, building systems that serve 100M+ tokens a day without catching fire.

I like systems where failure is a first-class feature: idempotency, retries, backpressure, graceful degradation, boring rollouts. I’m happiest when a graph goes down because we fixed it, not because we hid it.

I’ve had the chance to speak at DevOpsDays Bengaluru 2023 (because apparently I have opinions about API gateways now), and once won a hackathon by building a blockchain thing before it was cool. Also did a semester at UC Berkeley’s Sutardja Center, soaking up the Silicon Valley chaos.

Currently obsessed with: Go garbage collection tuning, making Kubernetes controllers boring, shaving milliseconds off inference latency, and turning real-world mess into reliable systems.

Bio (short version): I bridge AI research and production-scale systems. I talk about API gateways at conferences, design inference platforms that sling tokens at scale, and still get a kick out of hacking on protocol compilers. Community builder, systems tinkerer, and your friendly neighborhood latency slayer.

What I’ve Built

@ Lyric.tech

Senior Software Engineer, Core (Platform) — 2025–Present

  • Building core backend services in TypeScript for platform workflows and data APIs.
  • Getting a front-row seat to supply chain reality: messy inputs, real-world constraints, and a lot of “it depends”.
  • Design focus: correctness under change, evolvable schemas, and operational clarity (alerts that mean something).

@ Tune AI (formerly NimbleBox.ai)

Software Engineer — 2020–2025

  • Built a document processing system that chews through 100K documents daily with batch inference. Cut costs by 54% while keeping precision-recall at 95%.
  • Shipped a GenAI architecture for a UAE event management company — reduced their proposal creation time by 20% and automated 40% of their workflow.
  • Designed the gRPC backbone for a model serving system handling 100M tokens/day — latency dropped 40%.
  • Led the OpenTelemetry adoption because debugging distributed systems without traces is pain.
  • Built a reusable cron scheduler that killed 70% of boilerplate scheduling code. You’re welcome, future me.
  • Where it all started: ML pipeline orchestration, containerized model serving with FastAPI, and liberating models from Jupyter notebooks (not a fan).

Side Projects

A mix of things I’ve built, fixed, or accidentally broke:

  • Colbert AI — GPT-2 (345M) fine-tuning to mimic Stephen Colbert’s monologue voice using 500+ transcripts. Was doing LLM hijinks before it was fashionable. 😎
  • proto-opt-parse — Runtime parsing of Protobuf descriptor options in Go. Built to make gRPC tooling less guessy.
  • docker-hub-data — Data-X project at UC Berkeley: end-to-end pipeline to collect and analyze Docker Hub usage stats.
  • ovms-serving — OpenVINO Model Server tooling for inference workloads.
  • k8s-dashboard fix — Contributed a fix to the Kubernetes Dashboard. It was bugging me.
  • Also found a bug in the Go compiler’s release note item’s functionality. No big deal.

Notes, Talks, and Other Escapes

  • DevOpsDays Bengaluru 2023: APISIX as reverse proxy for high-volume inference traffic. Co-presented with Vinuja Khatode.
  • Visiting Scholar, UC Berkeley SCET (2019): Crashed the Data-X program on a $20K scholarship and built Docker Hub analytics pipelines.

Back in College

Next Tech Lab AP — Board Member

Helped scale from a scrappy group of <20 people to 150+ members. Lots of recruiting, mentoring, and figuring out how to build a culture around exploring cool stuff.

ACM Student Chapter — Founding Member

Started the chapter from scratch and grew it to 200+ members. Organized workshops, hack nights, and way too many pizza-fueled coding sessions.

Teaching Assistant — SRM University (2019–2020)

Ran tutorials for CS courses, built grading scripts in Python (because I’m lazy), and helped students debug their first tracebacks. Ensured a generation of students didn’t quit Python. You’re welcome.

Honors & Awards

Winner, IIIT-Delhi HACKIIITD (2017)

1st place out of 800+ teams. Built “Reliability of Media” — a blockchain-based consensus mechanism to combat misinformation. Incentivized honest votes with crypto rewards and locked consensus on-chain. Yes, blockchain. It was 2017, don’t judge.

Other Wins

  • Google Cloud Associate Cloud Engineer (Valid until 2026)
  • President’s Placement Award from SRM University
  • Grand Finalist — TCS IT Wiz 2013 (ask me tech trivia)

Tech I Actually Use

Daily DriversGo (system architecture, tooling), Python (MLOps, scripts), TypeScript (platform services)
InfrastructureKubernetes, Docker, AWS, GCP, Apache APISIX
Data StuffKafka, RabbitMQ, PostgreSQL, Redis
ObservabilityOpenTelemetry, distributed tracing, structured logging

Let’s Talk

Building something interesting? I’m always up for a conversation about distributed systems, backend architecture, or why gRPC is underrated.

Reading List

Trying to keep tabs on what I’m reading? I keep a running log on the Books page.