Dynamic Parser

home screen
preview your inputted data
add some target fields
it's parsing!
wow! results!
look at the raw results
look at the generated parsing code

Inspiration

We constantly scrape large amounts of raw, messy data at our company, BetterBasket, to get grocery store prices and competitor data. Writing parsers for each source is painful. Every time the data format changes, our code breaks. We needed a faster, smarter way to adapt.

What it does

Dynamic Parser turns AI into your data ally. You feed it raw JSON (or text) and describe what fields you need, and it instantly generates and runs a parser, returning clean, structured results in seconds.

How we built it

We used Daytona Sandboxes for safe code execution, OpenAI models for dynamic parser generation, and a simple Python interface that handles everything from field inference to result formatting.

Challenges we ran into

Dynamic code generation and sandbox execution handling
Managing huge JSON blobs with deeply nested structures

Accomplishments that we're proud of

We built a self-adaptive parsing system that actually works — no more endless json.loads debugging or broken scrapers. It saves hours of manual fixes and scales easily. We also pretty successfully created some preprocessing system on these large raw files so that the LLM knows what to look for.

What we learned

Building reliable AI agents requires balancing creativity with control. Giving models context about real-world data structures leads to surprisingly robust results.