3
Hey everyone! A few months ago I had to analyze a few thousand audio recordings to help identify customer support issues. I was able to get some crude high-level preliminary results with Python scripts invoking the LLM APIs, but they were very generic and unhelpful. Basic signals are easy to write, but tuning them and making them specific enough to ensure that no faint signals are missed is difficult. You need to iterate through the data with the initial prompt, splitting the data into different buckets, chaining another prompt for each bucket. Then you need to constantly review the raw data to fine-tune the prompt to get the desired results. There is no good user-facing tool for scaling thousands of rows of unstructured data analysis with LLMs. Cloud computing / agents accessing file systems are scratching the surface, but having a text-only UI is difficult, especially when you want to go back and adjust your research pipeline, limit to a specific subset of your data with filters like SQL, or manage a cost. Scaling past 100 files is not well supported. In-depth research is difficult to conduct and verify. I needed a mini data warehouse that would help me gain insights from my data, optimize costs with bulk LLM operations (through cost estimation and model selection), and let me browse and validate data in a user-friendly way, without the need to set up something like DataBricks. So, I made a folio. Folio is a free, native, macOS app for analyzing your unstructured data with LLMs. It’s a UI wrapper around a minimal data warehouse that lets users (and agents) perform LLM-based transformations on large unstructured data sets. All you need to get started is an AI API key and an account with modal.com Users bring their files into Folio which are then loaded into a table, where each row contains a Markdown representation of the file’s contents. Users can then run a large number of LLM operations on these files and create views using sql filters to reduce the scope of changes. Agents are first-class citizens and can plug into Folio to do most of the work for you. We integrate with modal.com as an execution engine to load the desktop for thousands of HTTP requests for OCR/audio transcription as well as AI APIs. A local orchestrator prepares jobs for modeling and then fans them once completed. Data is never stored anywhere, and is only transmitted through the AI ​​API provider and the user’s own modal infrastructure. Folio workspaces are multimodal (you can load different data types into the same workspace and move it through the same analysis pipeline) and they can support thousands of files. People today use Folio to: – Review customer support tickets/emails: narrow down bucket issues into different categories, categories of interest, and then act on that data by generating a response. – Extract detailed data from financial documents: load all data that can be found on a particular company, extract structured data such as revenue numbers and estimates. – Do literature reviews: There are many agents that help you load data from research paper repositories. Once that data is loaded into Folio, users can do deeper research on those files. – Criteria-based search: Create yes/no criteria such as “document contains data on XYZ”, “document mentions ABC”, “documented reference XYZ”. Companies like v7labs, Hebbia, Legora, Harvey have similar “tabular document review” features, but they are not scalable or compatible with external agents like Cloudcode. In addition, they require expensive enterprise agreements. I see Folio moving beyond data analysis to a great companion for agent tasks that require human-facing UI/UX, cost management and large amounts of data processing. Website: Github: X:
Looking forward to hearing what people think!