Trellis
AI-powered Snowflake for unstructured data
Trellis converts your unstructured data into SQL-compliant tables with a schema you define in natural language. With Trellis, you can now run SQL queries on complex data sources like financial documents, contracts, and emails. Our AI engine guarantees accurate schema and results. Leading enterprises use Trellis to: 1. Unlock hidden revenue in their customer data (e.g., Underwriting teams use Trellis to extract key features from transaction data and build better risk models.) 2. Supercharge RAG applications by enabling end-users to ask analytical questions not possible before with traditional Vector DB (e.g., what are the top three features that users are requesting) 3. Enrich their data warehouse with business-critical information (e.g., Retrieving detailed pricing and quantity information of products sold on competitor websites)
Jobs at Trellis
Active Founders
Mac Klinkachorn
Mac is the co-founder and CEO of Trellis. Previously, he worked at the Stanford AI lab on large multimodal models with the DoD and built ML infrastructure at Cresta, Moveworks, and Amazon. Mac started his first company at 15, building water leak detection systems, and grew it to six figures in ARR.
Jacky Lin
Jacky is a co-founder of Trellis and has taught hundreds of Stanford graduate students how to build, train, and deploy AI models in the Stanford School of Engineering & Graduate School of Business. Previously, he worked at Meta, the World Bank, and Wayfair.
Company Launches
TL;DR
Trellis extracts and transforms your unstructured data to SQL-compliant tables with schema you define with natural language. With Trellis, you can run SQL queries on your unstructured data sources like financial documents, contracts, customers, emails, etc.
✅ Try out the product live at https://demo.runtrellis.co
❌ Problem — Modern data warehouses are not designed for unstructured data
- For complex business and aggregation queries (i.e. what are the most common customer complaints or how do transaction categories change over time? ), RAG and traditional search techniques fall short.
- 80% of enterprise data is unstructured and arrives in many different formats. This makes it hard for applications and business decisions to be built reliably on top of that data.
- Building dedicated ML pipelines to extract features from unstructured data and performing inferences are time-consuming and hard to maintain. Teams are bogged down by handling edge cases.
🎉 Solution
Our AI engine combines LLMs, multimodal models, and database query engines to guarantee correct schema and accurate results across unstructured data sources.
- Defined your transformation and tasks with natural language. Trellis takes care of the rest.
- First-class support for PDF, HTML, and images with table transformation, context-aware chunking, and data normalization. Trellis robustly finds and aggregates all the data in situations where the smallest details matter.
- Real-time compute and feature store for all your analytics needs
Trellis in-action: Processing hundreds of pages for all common health insurance plans)
🏃🏽♂️ Leading Enterprises Use Trellis to:
- Supercharge RAG Applications: Enrich RAG pipelines and data warehouse by including business-critical information from unstructured data.
- Unlock Hidden Insights from Customer Calls and Emails: Process and query Terabytes of calls and emails to identify revenue opportunities, compliance risks, and user behavior over time
- Automate Contract Reviews and Underwriting: Extract key provisions from thousands of contracts, experiment with new features from financial data sources, and build better ML models from the newly transformed structured data
🙏🏼 Our Ask:
- Try out the live demo and join the waitlist for API access here.
- If you or someone you know is working with high volume of unstructured data (PDFs, calls, Log files), please reach out to us at founders@usetrellis.co.
- Feel free to pick a time here and we are more than happy to do a deep dive on a specific use case. We would love to serve as resources for building enterprise data infrastructure in the age of LLMs and beyond.
About Us:
Mac & Jacky are friends from the Stanford AI Lab, where we spent too much time indoors building language models and hoping that it would talk to us 🧠💻
Previously, Mac built LLM infra at Cresta and Moveworks where our ML pipeline processed Terabytes of chat logs and voice calls. Jacky taught AI classes at Stanford and worked at Meta on their real-time machine learning team handling complex model deployment challenges.
