PDFs are everywhere — invoices, reports, product catalogs, policy documents. While they’re great for sharing information, they’re not exactly easy to work with when you want to do something useful with the data inside.

This is where the combination of a large language model (LLM) and the GRID API comes in handy. LLMs can be used to extract structured values from a PDF, and the GRID spreadsheet engine can take those values and plug them directly into spreadsheet models that calculate, compare, and return accurate calculations — all via an API.

The spreadsheet engine plays a critical role in this process. While LLMs are impressive, they weren’t built for math. They regularly hallucinate numbers, miscalculate logic and confidently return incorrect results. That’s a deal-breaker when your AI needs to handle real business logic and provide outputs people can trust.

By pairing the language and document parsing capabilities of an LLM with the calculation power of the GRID spreadsheet engine, you get the best of both worlds: smart extraction and reliable math.

Together, this setup unlocks a wide range of use-cases: auto-calculating responses to form submissions, generating quotes or offers, or powering intelligent assistants that reply with real, accurate insights.

A practical example: From PDF to mortgage comparison tool

One use-case we’ve explored with a customer in the mortgage space (Win the House You Love) involves helping homebuyers understand their loan options more easily. Their team was looking to simplify the process of interpreting standard loan estimate PDFs.Imagine turning a static document like that into a live, interactive chat experience where key details are extracted using AI, the calculations are handled behind the scenes by a spreadsheet model, and the results are displayed in clear, interactive charts.Here’s what that flow could look like in practice:

1. Start with a PDF loan estimate

Loan estimates contain important numbers — interest rate, loan amount, loan term, taxes, insurance, fees — all of which shape the monthly payment and total cost of a loan. But the format they come in is static, typically a PDF, and not designed for easy data extraction or automation.

2. Extract the relevant values using an LLM

An LLM reads the PDF and pulls out the necessary values and labels. LLMs are well-suited for this task as they are especially good at understanding semi-structured documents like these, where the data is consistent but the formatting can vary slightly between sources. They’re able to interpret headings, fields, and context — pulling out standard data like loan amount, term length, and escrow payments without needing custom logic.

3. Send the values to a spreadsheet model via the GRID API

The extracted values are passed into a spreadsheet model hosted in GRID which already contains the formulas and logic to calculate monthly payments, compare loan types, and summarize costs.

This is where business logic lives — in a format many teams already know and trust. With the GRID API, there’s no need to recode this logic into a custom backend. The spreadsheet is the backend.

4. Power the chat with calculated spreadsheet outputs

The spreadsheet runs in the background, and the results — monthly payment, total loan cost, comparison metrics or even visualizations like charts — are delivered to users through the interface.

The GRID API acts as the connective layer, ensuring each response is accurate, verifiable, and reflects the latest inputs.

As users interact with the chat, their inputs are sent to the spreadsheet via the API, triggering live calculations. These are done dynamically — the original model remains untouched. Once connected, a spreadsheet is effectively transformed into a scalable web service that can respond to queries in real time, run calculations on demand, and return results, while preserving the original logic and data.

This makes it easy to compare options, test different scenarios, or support multiple users without risking unwanted edits to the source file.

Why it works

This setup has a few major advantages:

• Flexible data extraction: The LLM handles variability in formats — pulling out key values like interest rate, down payment, and insurance even when formatting differs between documents.

• Accurate, verifiable calculations: All the logic lives in a spreadsheet — where it likely already exists. By running extracted values through a spreadsheet engine, you get accurate, verifiable results. With the GRID API, your spreadsheet becomes a live backend that recalculates on the fly as users interact.

• Everything connected by API: The workflow is powered by simple, flexible API calls. That makes it easy to plug into a chat assistant, an internal tool, or even a public-facing application.

It’s a great example of how modern language models and spreadsheet logic can work together — turning static content into actionable insights grounded in accurate calculations.

Want to try it yourself?

You can sign up and start using the GRID API to build workflows like this with your own spreadsheet data and models. If you’re working with LLMs, PDFs, or any kind of data extraction, it’s a fast and reliable way to automate real business logic — without rebuilding what already works.

Leave the language to LLMs and the numbers to us.

👉 Sign up

📘 API Documentation