Divyanshu Goyal

Marketing Quality Vision Language Model

Training vision-language models to solve the complex problem of evaluating marketing image quality at scale. Designed and implemented a novel observable injection technique that fundamentally changes how VLMs reason about image quality by augmenting prompts with quantitative metrics (sharpness, color balance, exposure, lighting distribution). This approach bridges the gap between learned representations and interpretable measurements, enabling the model to ground its assessments in concrete image properties. Tackled the challenging problem of Photoshop edit evaluation, training the model to understand subtle quality differences across multiple dimensions. Developed MQCore, a comprehensive evaluation score inspired by DCLM CORE, to rigorously benchmark model performance on marketing quality tasks. Built the complete training pipeline including multi-GPU distributed training with LoRA, synthetic conversation generation for diverse training data, and evaluation frameworks combining both DCLM CORE and custom MQCore benchmarks.

Adobe GenStudio for Performance Marketing

Led the development of the foundational prototype and presented to Adobe executive leadership, driving the strategic decision to launch GenStudio in October 2024. This generative AI-first platform addresses a critical market need where content demand is expected to grow 5x by 2026, enabling marketing teams to create on-brand campaign content at scale. Built with Adobe Firefly for image generation and enterprise LLMs for copy generation, the platform integrates with major advertising platforms (Google Campaign Manager 360, Meta, Microsoft Advertising, Snap, TikTok) and provides AI-powered brand validation. Now serving enterprise marketing teams across major brands.

Curiosity-Driven LLM-as-a-judge for Personalized Creative Judgment

Vanya Bannihatti Kumar, Divyanshu Goyal, Akhil Eppa, Neel Bhandari

Modern large language models (LLMs) excel at objective tasks such as evaluating mathematical reasoning and factual accuracy, yet they falter when faced with the nuanced, subjective nature of assessing creativity. In this work, we propose a novel curiosity-driven LLM-as-a-judge for evaluating creative writing, which is personalized to each individual's creative judgments.

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

Systems and techniques to efficiently serve fine-tuned models by dynamically loading parameter-efficient layers into pre-loaded base models. This approach drastically minimizes network data transfer costs, container startup time, and provides a highly efficient, scalable system for serving specialized machine learning models in production environments.

Brand-Aligned Marketing Content Generation Using Structured Brand Data and Generative Models

Mayank Anand, Jose Mathew, Divyanshu Goyal

Systems and methods employ generative models to transform unstructured brand information into structured brand data and generate brand-aligned marketing content. The system uses natural language processing and machine learning to extract brand DNA elements, generate confidence scores for structured components, and create marketing content that accurately reflects an entity's brand identity across various channels.