Divyanshu Goyal
Applied Scientist at Adobe
Applied Scientist at Adobe specializing in cutting-edge machine learning solutions. I hold an M.S. in Computer Science from Georgia Tech with a focus on Machine Learning. Passionate about training Large Language Models and Vision-Language Models, I thrive on solving real-world challenges that drive meaningful impact.
About Me
Education
M.S. in Computer Science
Georgia Institute of Technology
Machine Learning
B.E. in Computer Science
Birla Institute of Technology and Sciences, Pilani
M.Sc. in Mathematics
Birla Institute of Technology and Sciences, Pilani
Expertise
Tools & Languages
Experience
Applied Scientist
—Adobe, San Jose, CAJune 2021 — Present
Vision-Language Model Training & Evaluation
- Designed and trained a 13B-parameter VLM on 2M image-text pairs across 8x H100 GPUs, using novel observable injection techniques.
- Fine-tuned CLIP and LLaVA-style models, boosting F1 score by 20% in multi-label classification tasks.
- Developed MQCore evaluation benchmark inspired by DCLM CORE for domain-specific quality assessment.
LLM Alignment & Multimodal Reasoning
- Created curiosity-driven LLM-as-a-judge framework for personalized creative evaluation (arXiv 2025).
- Pioneered creativity modeling with Bayesian surprise, achieving 23% F1 and 38% correlation improvements.
- Leading development of multimodal agentic systems for Photoshop editing workflows.
Production AI & Patented Inventions
- Led development of Adobe GenStudio for Performance Marketing (Oct 2024) — now serving Fortune 500 clients including Microsoft and AT&T.
- Invented cross-cultural image adaptation pipeline using VLM grounding + diffusion models (US Patent 2026).
- Designed constrained machine translation system with facet-level adherence scoring (US Patent 2026).
Machine Learning Engineer
—Adobe, Bangalore, IndiaJuly 2016 — August 2019
ML Infrastructure & Serving
- Co-engineered Adobe's ML inference platform — 3,500 QPS/node at 0.3–0.6ms latency (p99).
- Designed GC-free async event pipeline using LMAX Disruptor, eliminating back-pressure under peak load.
- Built custom model lifecycle management platform — versioned deployment, monitoring, and staged rollout across environments.
Projects
Marketing Quality Vision Language Model
Built a complete MLOps system for training vision-language models to assess marketing content quality, featuring novel observable injection and Photoshop edit evaluation capabilities.
Adobe GenStudio for Performance Marketing
Created the first demo and pitched to Adobe leadership, leading to the launch of a generative AI-first platform that revolutionizes marketing content creation at scale.
Publications & Patents
Curiosity-Driven LLM-as-a-judge for Personalized Creative Judgment
Vanya Bannihatti Kumar, Divyanshu Goyal, Akhil Eppa, Neel Bhandari
Systems and Methods to Provide Parameter-Efficient Fine-Tuned Models
Yuanyou Wang, Naveen Vangala, Mayank Anand, Kunal Kumar Jain, Jose Mathew, Eapen Jose, Divyanshu Goyal, Asmita Chihnara, Arif Abdullah, Anand Dantu
Brand-Aligned Marketing Content Generation Using Structured Brand Data and Generative Models
Mayank Anand, Jose Mathew, Divyanshu Goyal
Cultural Adaptation of Images Using Generative Artificial Intelligence
Akhil Eppa, Mayank Anand, Divyanshu Goyal
Guiding Language Translation with Translation Documents Using Machine Learning
Divyanshu Goyal, Akhil Eppa, Mayank Anand
Writing
A Lazy Engineer's Guide to Scaling Transformers
How to compress transformer configuration into a single integer — depth. By deriving architecture dimensions (n_layer, n_embed, n_head), batch size, learning rate, weight decay, and training horizon from one number, scaling sweeps across multiple model sizes become trivial and error-free.
Going Fast: Every Optimization That Made LLM Training Fly
Six concrete optimizations — TF32, BF16 mixed precision, torch.compile, Flash Attention, parallel DataLoaders, and pinned memory transfers — that collectively drove a significant jump in Model Flop Utilization (MFU) while training VibeNanoChat, a GPT-2 scale LLM. Each technique is explained with real code and the hardware-level reasoning behind it.
Engineering High-Throughput, Low-Latency Machine Learning Services
A deep dive into building Adobe's User Response Prediction Service that processes 3,500 requests per second with sub-5ms latency. Covers code optimization, garbage collection tuning, event generation, and performance benchmarking techniques for high-performance ML services.