Ishan Upadhyay

Member of Technical Staff at Adobe

Ishan Upadhyay

I build cross-platform software and developer tools for Creative Cloud Desktop. My work spans C++ systems, applied AI, and research in model evaluation and computational biology.

Get in touch Resume (PDF) GitHub
Ishan Upadhyay

Experience

Industry & product engineering
Adobe

May 2025 — present

Member of Technical Staff

Creative Cloud Desktop · Windows & macOS

I work on C++ platform layers, React interfaces, and developer tooling for Creative Cloud Desktop, a product serving 50M+ users.

  • Built Patrol, a multi-agent system for production debugging and draft pull requests, with human approval required for write actions.
  • Delivered an Adobe Express Photos cross-platform integration prototype in under five weeks.
  • Implemented incremental feature-flag caching, reducing redundant network calls and saving approximately $200K in service costs, with a rollback mechanism.
  • Named inventor on a browser-to-desktop communication patent application, currently under filing.

Recognition Adobe Special Contribution Award for work including Patrol, cross-platform integration, and the patent application.

Adobe

May — July 2024

Product Engineering Intern

Creative Cloud Desktop · Recommendation systems

Developed app-recommendation models using LSTMs, collaborative filtering, and T5. The T5 next-best-app model achieved 80% top-5 accuracy; LSTM models reached 54% for Express and 56% for Firefly.

Deployed the models with FastAPI and built a Streamlit interface for interactive evaluation.

Selected work

Systems I have built

Adobe · Developer tooling

Patrol

I solo-built a debugging agent that investigates production issues, assembles evidence-backed root-cause analyses, and drafts pull requests.

100+production issues resolved
60–70% end-to-end resolution rate
Technical contribution
  • Designed a supervisor that coordinates parallel specialist agents.
  • Kept write actions behind human approval rather than giving the system unrestricted access.
  • Added a memory write-guard to validate corrections before persisting them.

Multi-agent systems · C++ · TypeScript

Independent project · Document processing

RadiologyAI

A prototype for extracting and searching information in radiology reports. I integrated LandingAI’s document extraction with Pathway’s streaming document pipeline.

Incoming PDFs are parsed into structured fields and indexed for retrieval through REST and MCP interfaces.

Technical contribution
  • Implemented a custom Pathway parser for LandingAI’s Agentic Document Extraction API.
  • Defined a schema for study types, findings, impressions, and other report metadata.
  • Exposed document retrieval and patient-ID search through the application’s interfaces.
Source code Read the article

Python · Pathway · LandingAI · MCP

Earlier projects

UniTrain

Contributed to an open-source Python library for classification, segmentation, GAN training, and FastAPI model inference.

2023 · Hello FOSS, IIT Bombay

IIT Bombay Mars Rover

Built image super-resolution and rock-classification models for the rover’s perception subsystem. The team won Best Autonomous at IRC and Best Navigation at ERC.

2022–2023 · Computer vision

Lyric generation with minGPT

Trained a lyric-generation model and compared single- and multi-threaded inference through a local FastAPI server.

2023 · Independent project

Writing

Technical articles & project write-ups

Research

Publications & collaborations
ACL 2025Co-author

GRACE: A Granular Benchmark for Evaluating Model Calibration against Human Calibration

Yoo Yeon Sung, Eve Fleisig, Yu Hou, Ishan Upadhyay, Jordan Boyd-Graber

With Prof. Jordan Boyd-Graber at the University of Maryland, I built the question-authoring platform used by 100+ writers, including AI feedback, Google OAuth, and collaborative drafting through Google Docs.

Paper PDF
Molecular & Cellular Proteomics2025 · Co-author

AlphaCross-XL: A Seamless Tool for Automated and Proteome-Scale Mapping of Cross-Linked Peptides Onto 3D Protein Structures

Worked with Prof. Sanjeeva Srivastava on a React/Node.js interface for protein cross-linking analysis, integrating ProDy, PyMOL, and pyfastx, and improving computation through parallelization.

Paper Code
ICML 2026 workshopContributor

QANTA 2026: Efficient, Incremental Multimodal Question Answering

Lead the Question Authoring Interface for human–AI co-authoring of text-and-image quizbowl questions, with Prof. Jordan Boyd-Graber and collaborators at UMD and UC Berkeley. I also contribute to workshop and competition organization.

Project Workshop

About

Indian Institute of
Technology Bombay

B.Tech. Aerospace Engineering
Class of 2025 · CPI 8.08 / 10

I graduated from IIT Bombay in 2025 and joined Adobe as a Member of Technical Staff. Alongside product engineering, I have contributed to research in model calibration, multimodal question answering, and protein-structure analysis.

At IIT Bombay, I served as Institute Manager for Web and Coding in 2023–24. I led a team of 10, helped coordinate 70+ summer projects, and helped teach Flutter, NLP, and DSA courses reaching 3,000+ students. I also mentored 24 freshmen as a CS101 teaching assistant.

Languages
Python, C++, TypeScript, JavaScript, C
Frameworks
React, Node.js, FastAPI, PyTorch, TensorFlow, CEF

Contact

Email is the best way to reach me.

LinkedIn GitHub Resume (PDF)