Andera × Lyra: Deep in the Document Layer, Through a $37M Raise
How Lyra worked behind the scenes with Andera to solve one of the less glamorous parts of building AI: transforming a 40MB audit workbook into structured data an agent can read, evaluate, and improve against.
.png)
About Andera
Andera is an AI-native platform built for internal audit and SOX compliance testing. It takes evidence exactly as it comes in, whether that's PDFs, screenshots, ledger sheets, or massive Excel workbooks, runs the required control tests, and produces workpapers in the same format audit teams already know and use.
Founded in 2024 by Aryo Patel and Tinah Hong, based in San Francisco, out of PearX Summer 2024. In June 2026, Andera raised a $37M Series A led by Lightspeed Venture Partners, with Bain Capital Ventures, A* and Pear joining.
The brief
Andera came to Lyra in March 2025 for engineering help on the layer where evidence enters the product. Lyra started on PDF processing in the first stretch of the engagement, then moved into Excel, which is where the real problems live.
What we shipped
Lyra worked inside the ingestion and evaluation layers, where most of the difficulty in this product sits:
- Document extraction. Broader support for the files clients actually send, so less evidence needs manual handling.
- Excel at scale. Large, heavy workbooks made fast to upload, read and process.
- Spreadsheet fidelity. Rebuilt workbooks that match what the client uploaded, structure and formatting intact.
- Evaluation tooling. Testing that judges whether the AI completed an audit task correctly, and feeds that back into the product.
- Agent and prompt work. Sharper feedback from the AI, and the logic to pinpoint what caused a poor result.
The result
- Hundreds of Excel workbooks handled across the engagement.
- A multi-workbook processing workflow cut by 55%, from minutes to seconds.
- Parts of the workbook data extraction process cut by around 64%.
- Trace processing cut by around 77%, taking one workflow from minutes to seconds.
The takeaway
Every product has a layer underneath the parts customers see, and it usually decides whether the product holds up in the real world. It is slow, specialist work, and it eats the engineering time a founding team would rather spend on the product itself. Andera kept its own team on the product and handed that layer to an engineer who went deep and stayed in it.
That is the model: your team on what customers see, a Lyra engineer owning the part underneath that has to hold.
Ready to build with us?
Whether you're filling an engineering gap or scaling alongside your senior team, Lyra embeds with you from day one.
.webp)
