SG Shaddy Garg

Machine Learning Engineer · Systems Builder

Building AI systems that survive the real world.

I’m Shaddy Garg. I design and build production systems where machine learning, distributed infrastructure, and enterprise knowledge meet.

Explore the systems ↓ Google Scholar ↗

Featured engineering systems

From research idea to operating system.

The model is rarely the hardest part. These case studies focus on architecture, failure modes, evaluation, and the engineering decisions that make AI useful at scale.

01 / Enterprise AIProduction system

Adobe Experience Platform

Enterprise Conversational AI Assistant

A modular assistant platform that combines structured-data reasoning, retrieval, evaluation, and auditable answers across enterprise knowledge.

Problem
Trustworthy answers across fragmented enterprise data
Scale
Multiple domains, data types, and evolving use cases
Core decision
Route by intent; ground every answer in evidence
02 / Data systemsResearch → resilience

EvoSchema · PVLDB 2025

Schema-Resilient Text-to-SQL

A benchmark and augmentation pipeline for keeping natural-language database interfaces reliable as schemas evolve underneath them.

Problem
Models silently break after real schema changes
Challenge
Preserve intent across structural evolution
Tradeoff
Robustness without retraining for every migration
03 / ReliabilityEnterprise cloud

Outage-Watch + ESRO

Proactive Cloud Reliability

Forecasting, causal diagnosis, and experience-assisted remediation for complex services where failures are rare, expensive, and difficult to define.

Impact
Up to 88% lower mean time to detection
Architecture
Forecasting + causal graph + experience graph
Failure mode
Rare-event drift and alert noise

Numbers that matter

0Peer-reviewed publications
0Patented inventions
0Years building at Adobe
Prod.AI systems in enterprise use

Currently exploring

Questions with energy.

01

Enterprise AI

Systems that reason within policy, context, and changing organizations.

02

Agents & memory

Long-running workflows with useful state, clear boundaries, and observable decisions.

03

Knowledge systems

Graphs and semantic retrieval for provenance-aware, multi-hop reasoning.

04

Evaluation & causality

Connecting human judgment, production signals, and explanations engineers can act on.

Publications

Ideas, in print.

Peer-reviewed work spanning AI, databases, software engineering, web systems, and visualization.

2026

SIGMOD Companion

Building, Serving, and Growing a Conversational AI Assistant for Enterprise

S. Bhatia, V. Bhatia, U. Bhattacharya, et al., including Shaddy Garg
2025

PVLDB

EvoSchema: Towards Text-to-SQL Robustness Against Schema Evolution

T. Zhang, K. Qian, S. Sahai, Y. Tian, Shaddy Garg, H. Sun, Y. Li
2023

AAAI

Reinforced Approximate Exploratory Data Analysis

Shaddy Garg, S. Mitra, T. Yu, Y. Gadhia, A. Kashettiwar
2023

The Web Conference

CausIL: Causal Graph for Instance Level Microservice Data

S. Chakraborty, Shaddy Garg, S. Agarwal, A. Chauhan, S. Saini
2023

ASE

ESRO: Experience Assisted Service Reliability Against Outages

S. Chakraborty, S. Agarwal, Shaddy Garg, A. Sethia, et al.
2023

ESEC/FSE

Outage-Watch: Early Prediction of Outages Using Extreme Event Regularizer

S. Agarwal, S. Chakraborty, Shaddy Garg, S. Bisht, et al.
2023

SIGMOD Companion

Fast Natural Language Based Data Exploration with Samples

S. Agarwal, G. Chan, Shaddy Garg, T. Yu, S. Mitra
2022

VISSOFT

ViSRE: A Unified Visual Analysis Dashboard for Proactive Cloud Outage Management

P. Kayongo, J. Hoffswell, S. Saini, Shaddy Garg, et al.
2020

arXiv

SPLIT: Inferring Unobserved Event Probabilities for Disentangling Brand-Customer Interactions

A. Chauhan, A. Anand, Shaddy Garg, S. Dhamnani, S. Saini
View live citation record on Google Scholar ↗

How I think

Build for the system,
not the demo.

I enjoy problems where machine learning meets distributed systems.
I care less about benchmark scores and more about systems that remain reliable in production.
Research becomes valuable when it survives contact with real users.
I like turning difficult infrastructure into intuitive developer experiences.
View résumé ↗

Journey

  1. 01
    CybersecurityLearn to think in failure modes
  2. 02
    ResearchForm precise questions and test assumptions
  3. 03
    Distributed systemsDesign for partial failure and changing state
  4. 04
    Machine learningBuild with uncertainty, data, and feedback
  5. 05
    Enterprise AIEarn trust at organizational scale
  6. 06
    Knowledge systemsConnect reasoning to evidence and context

Selected patents

Applied research translated into protected technology at Adobe.

Recognition

Built to explore.
Driven to compete.

01

Adobe

Outstanding Young Engineer Award

2023
02

Google Summer of Code

OWASP Nettacker contributor

2018
03

AISS Security Hackathon

First place

2018
04

Technex CTF

Global rank #2

2019
05

Brainwaves Cyber Security Challenge

Runner-up

2019
06

BSides Delhi CTF

Second runner-up

2018
07

National recognition

KVPY Scholar · AIR 963, JEE Advanced

2016

Build in public

The workbench is open.

Current experiments, unfinished questions, and notes moving from idea to artifact.

Exploring now

Memory architectures for enterprise agents

Separating durable facts, episodic context, and short-lived working state.

Experiment

Evaluation harness for grounded answers

Connecting human review, claim-level attribution, and production telemetry.

Upcoming note

Retrieval beyond RAG

A decision framework for lexical, vector, graph, and hybrid retrieval.

Open question

How should agents know when to stop?

Bounded autonomy, uncertainty, cost, and explicit handoff as system primitives.

Outside engineering

🏋️ Functional fitness✈️ Travel🍜 Food 📷 Photography🎵 Learning

Start a conversation

Interesting systems.
Hard problems.

I enjoy conversations about distributed systems, enterprise AI, databases, evaluation, and ambitious engineering problems.

Send me a note ↗

System case study

Source / paper ↗