Skip to content
Sai Vivek Peddi

Engineer,
researcher,
co-founder

Grew up in Hyderabad, now in Mountain View. I make machine learning systems faster, cheaper and more trustworthy, and I co-founded mlpal.

I'm Sai. Close friends call me Peddi.

I build AI systems
and great dosa.

Sai Vivek Peddi, smiling, in a white sweater, sitting in front of window blinds

I grew up in Hyderabad, studied my way into BITS Pilani, and have spent the years since making machine learning systems faster, cheaper and more trustworthy.

Today I’m co-founding mlpal with Dipak Ghosal and Prem Jain. Off the clock you’ll find me on a badminton court, bowling a few overs, behind a camera, baking bread, or halfway through a book.

The story so far

Hyderabad → Mountain View

01 / Hyderabad

Origin

Grew up here and studied my way into BITS Pilani.

Home.

02 / BITS Pilani

2015

Electronics and communication engineering. Signals first, then machine learning.

B.E. (Hons.), Hyderabad campus.

03 / IIIT Hyderabad

2018

Research assistant in the speech processing lab. My first real taste of ML research.

ML and speech processing.

04 / KFintech

2019

Architected Digix, the reporting platform for 18 asset managers.

$137B AUM, 100M investor accounts. Led five engineers; Spark pipelines cut query times 40×.

05 / UC Davis

2021

M.S. in Computer Science with a 4.0, and four terms as a TA.

Research with Prof. Dipak Ghosal on GPUs, networks and distributed training.

06 / JPMorgan Chase

2023

Senior ML engineer. Brought coding agents to more than a thousand engineers.

Multi-agent systems in production, 1M+ requests a day at 99.9% uptime.

07 / mlpal

2026

Co-founded mlpal with Dipak Ghosal and Prem Jain, to give companies ownership of the loop around the model.

Gateway, harness and memory. Open source, research-led.

Things I’ve built

(08)
CUDA

Sparse attention kernels

2022

Custom kernels for ViT self-attention: 100× faster at 85% sparsity, 97.5% of the GPU busy.

Visualization

Volumes, super-resolved

2021

Interactive isosurface rendering of CT scans and supernovae, sharpened with deep learning.

Systems

Latency-aware Kubernetes

2022

An EDF scheduler on Prometheus telemetry that holds QoE targets at 75% utilization.

Off the clock

Courts, kitchen, headphones

Badminton

Badminton, any day.

Tap the shuttle. Rally 0 · Best 0

Cricket

Dhoni

7

  • All-rounder: I bat and I bowl.
  • Big-time Dhoni fan.

Kitchen

Naan with curry and lime, plated on a dark tableA plated dish with greens beside a glass of red wine
  • I make a great dosa.
  • I bake my own bread.
  • Two new recipe books, working through them.

Music

“Music has no language and no boundaries.”

It changes from time to time. Right now: indie from India and Pakistan, and lately some Italian.

On repeat

Abhijeet Srivastava

Books

Just finished

The Secret of Secrets

Dan Brown

Photography

Through my lens

Skies, streets and the occasional bird. Shot on a Canon, graded by hand, and flown by drone when the view calls for it.

Black-necked stilt
Red sky, Davis Airport
Center line
Giant sequoias, looking up
After the rain
Sequoia rings
Bike in the rain
Milky Way
Half Dome
Pacific coast
Full moon
Dusk, Davis Airport
On the wire
Winter tree at sunset

More on Instagram @_forgotten_chornicles

Community

BITS Alumni Association, Bay Area

11,000

BITSians in the Bay Area

I co-lead the BITS alumni association here as its COO. We're a registered nonprofit, and we run events all year to bring this community together.

Lots of events, one goal: keep the BITS spirit alive a long way from campus.

Research & teaching

All mlpal research (08)

Writing

All notes (01)
The model is rented. The loop should be yours.Same model, same answers, very different bills. Why the harness around the model is where cost, reliability and learning actually live.3 min read
Stay hungry, stay foolish.
Steve Jobs