Case Study - Acrobat Assistant

Worked on fine tuning language models for document intelligence lab

Fine Tuning Models


Adobe Atlas is revolutionizing the research landscape by giving scholars the tools to dynamically design studies around PDF content. Using Python scripting, researchers can produce adaptable JSON study designs and then instantly pair them with PDFs for a seamless study formulation process. When these PDFs are uploaded to the Atlas site, they're transformed into engaging, survey-like studies designed for user interaction. This interface captures user insights, with studies yielding actionable data output in JSON format.

Post data collection, the platform excels in extracting valuable text and imagery from user selections, streamlining data preparation for ML models. Furthermore, the gathered data is harnessed to refine a proprietary language model, leading to the creation of "Acrobat Assistant", a next-generation tool whose details are eagerly awaited for public release.

What we did

  • full-stack development
  • model intergration
  • natural language processing

Adobe Atlas revolutionizes the research landscape by empowering scholars with the tools to dynamically design studies centered around PDF content.

Naman Kapasi
Research Scientist at Adobe

More case studies

Lofi Bytes: AI-Generated Lofi Music Web App

Lofi bytes is a web app designed to allow users to create their own AI-generated lofi music using transformers and LSTMS.

Read more

EU Parliament Text Sentiment Analysis

Sentiment analysis and classification on transcripts from European Union Parliament sessions.

Read more

Tell us about your project